BYBEE, Circuit Judge

On Petition for Review of an Order of the Board of Immigration Appeals, Agency No. AXXX-XXX-XXX Aaron Korthuis (argued), Matt Adams, and Leila Kang, Northwest Immigrant Rights Project, Seattle, Washington; Alison Hollinz and Christopher P. Stanislowski, Northwest Immigrant Rights Project, Tacoma, Seattle; for Petitioner. Andrew C. MacLachlan (argued), Bryan S. Beier, and Zoe J. Heller, Senior Litigation Counsel; Jaclyn E. Shea, Trial Attorney; John W. Blakeley, Assistant Director; Joseph H. Hunt, Assistant Attorney General; Brian M. Boynton, Principal Deputy Assistant Attorney General; Office of Immigration Litigation, Civil Division, United States Department of Justice, Washington, D.C.; for Respondent. Kara Hartzler and Vincent Brunkow, Assistant Federal Public Defenders, Federal Defenders of San Diego Inc., San Diego, California, for Amicus Curiae Ninth Circuit Federal Public and Community Defenders. Before: Mary H. Murguia, Chief Judge, and Sidney R. Thomas, M. Margaret McKeown, Jay S. Bybee, Consuelo M. Callahan, Ryan D. Nelson, Eric D. Miller, Bridget S. Bade, Daniel P. Collins, Kenneth K. Lee and Lawrence VanDyke, Circuit Judges. Opinion by Judge Bybee; Partial Concurrence and Partial Dissent by Judge Collins; Concurrence in the Judgment by Judge Callahan; Dissent by Judge McKeown; Dissent by Judge VanDyke

OPINION

BYBEE, Circuit Judge, announced the judgment of the court, in which CALLAHAN, R. NELSON, MILLER, BADE, COLLINS, and LEE, Circuit Judges, concur; and delivered an opinion, in which R. NELSON, MILLER, and LEE, Circuit Judges, concur, and in which CALLAHAN and BADE, Circuit Judges, concur as to Subsections B and C of Part IV and Part V:

Petitioner McKenzy Alii Alfred is a non-citizen who was convicted of robbery in Washington. After Alfred finished a fifteen-month prison sentence, the government initiated removal proceedings against him. The Board of Immigration Appeals ("BIA") found Alfred removable for having committed an aggravated felony.

Alfred petitioned for review, and the panel granted relief based on our prior holding in United States v. Valdivia-Flores, 876 F.3d 1201 (9th Cir. 2017). In that case we held that Washington's accomplice liability statute was "implicit . . . in every criminal charge" and categorically "broader than its federal analogue." Id. at 1207, 1208. The government petitioned for rehearing en banc, arguing that Valdivia-Flores and the panel decision were incorrect and in direct conflict with a decision of the Eleventh Circuit, Bourtzakis v. United States Attorney General, 940 F.3d 616 (11th Cir. 2019). We granted the petition for rehearing en banc; we now overrule Valdivia-Flores and deny Alfred's petition for review.

I. BACKGROUND

A. Factual History

Alfred is a citizen of the Republic of Palau. He lawfully entered the United States in 2011 as a noncitizen pursuant to the Compact of Free Association. On February 20, 2018, Alfred went on a one-man crime spree in which he attempted to rob a credit union, successfully robbed a coffee stand, and attempted to steal a car. Police apprehended him shortly thereafter.

After his indictment, Alfred pleaded guilty to one count of second-degree robbery and two counts of attempted second-degree robbery. He was given a fifteen-month sentence for each count, with the sentences to be served concurrently. B. Procedural History

Shortly after Alfred was released from prison, the Department of Homeland Security ("DHS") initiated removal proceedings against him. The Notice to Appear charged Alfred as removable for having been convicted of an aggravated felony, a crime of moral turpitude, and an aggravated felony crime of violence.

Before the immigration judge ("IJ"), Alfred argued that the statute of conviction for robbery in Washington, Wash. Rev. Code § 9A.56.190, was overbroad under the categorical approach. The IJ rejected this argument, relying on United States v. Alvarado-Pineda, 774 F.3d 1198, 1203 (9th Cir. 2014), in which we held that "a conviction for Washington second-degree robbery, where accompanied by a sentence of at least one year, qualifies as an 'aggravated felony' under 8 U.S.C. § 1101(a)(43)(G)." As a result, the IJ ordered Alfred's removal, for commission of both an aggravated felony and a crime involving moral turpitude. The Board of Immigration Appeals affirmed the IJ's finding that Alfred had committed an aggravated felony but declined to address whether Alfred was removable on other grounds.

Alfred petitioned for review. The panel found that our precedent, namely Valdivia-Flores, compelled it to grant Alfred's petition. See Alfred v. Garland, 13 F.4th 980, 987 (9th Cir. 2021) ("[I]n this case, our analysis begins and ends with Valdivia-Flores."). Sitting by designation, Judge Morrison C. England, who authored the majority opinion, concurred specially. Judge Rawlinson concurred in the result. In Valdivia-Flores, we made two key holdings: (1) because aiding and abetting liability is implicit in every criminal charge, a state's aiding and abetting statute must be folded into our analysis under the categorical approach, and (2) Washington's aiding and abetting statute is broader than its generic equivalent. Valdivia-Flores, 876 F.3d at 1207-09. Based on these holdings, the Alfred panel found that Washington's robbery statute is a categorical mismatch to its generic equivalent. 13 F.4th at 986-87. Accordingly, the panel found that second-degree robbery was not an aggravated felony for removal purposes and that, as a result, Alfred was not removable.

The government petitioned for rehearing en banc, asking us to overrule Valdivia-Flores. We granted the government's petition. Alfred v. Garland, 35 F.4th 1218 (9th Cir. 2022).

II. JURISDICTION AND STANDARD OF REVIEW

We have jurisdiction under 8 U.S.C. § 1252, and we review the BIA's determinations of law de novo. Vitug v. Holder, 723 F.3d 1056, 1062 (9th Cir. 2013). "[W]hether an offense is an aggravated felony for removal purposes is a question of law." Chavez-Solis v. Lynch, 803 F.3d 1004, 1006 (9th Cir. 2015) (quoting Chuen Piu Kwong v. Holder, 671 F.3d 872, 876 (9th Cir. 2011)).

III. THE CATEGORICAL APPROACH

Under the Immigration and Nationality Act ("INA"), the government may order the removal of noncitizens who have committed crimes classified as "aggravated felonies." 8 U.S.C. § 1227(a)(2)(A)(iii); see also Moncrieffe v. Holder, 569 U.S. 184, 187, 133 S.Ct. 1678, 185 L.Ed.2d 727 (2013). The INA defines aggravated felonies with a list of offenses, including "a theft offense (including receipt of stolen property) or burglary offense for which the term of imprisonment is at least one year." 8 U.S.C. § 1101(a)(43)(G).

The INA specifies "conviction, not conduct, as the trigger for immigration consequences." Mellouli v. Lynch, 575 U.S. 798, 806, 135 S.Ct. 1980, 192 L.Ed.2d 60 (2015). As a result, the Supreme Court has instructed us to "employ a categorical approach by looking to the statute . . . of conviction, rather than to the specific facts underlying the crime." Esquivel-Quintana v. Sessions, 581 U.S. 385, 389, 137 S.Ct. 1562, 198 L.Ed.2d 22 (2017) (quoting Kawashima v. Holder, 565 U.S. 478, 483, 132 S.Ct. 1166, 182 L.Ed.2d 1 (2012)). The Court has applied this categorical approach in "several statutory contexts," including the Armed Career Criminal Act (ACCA). Borden v. United States, — U.S. —, 141 S. Ct. 1817, 1822, 210 L.Ed.2d 63 (2021) (plurality opinion).

The categorical approach asks us to determine "whether 'the state statute defining the crime of conviction' categorically fits within the 'generic' federal definition of a corresponding aggravated felony." Moncrieffe, 569 U.S. at 190, 133 S.Ct. 1678 (quoting Gonzales v. Duenas-Alvarez, 549 U.S. 183, 185-87, 127 S.Ct. 815, 166 L.Ed.2d 683 (2007)). To do so, we "compare the elements of the crime of conviction with the elements of the 'generic' version of the listed offense—i.e., the offense as commonly understood." Mathis v. United States, 579 U.S. 500, 503, 136 S.Ct. 2243, 195 L.Ed.2d 604 (2016). "[I]f the statute sweeps more broadly than the generic crime, a conviction under that law cannot count as an [aggravated felony], even if the defendant actually committed the offense in its generic form." Descamps v. United States, 570 U.S. 254, 261, 133 S.Ct. 2276, 186 L.Ed.2d 438 (2013).

In the case before us, the BIA found that Alfred had been convicted of a generic theft offense under 8 U.S.C. § 1101(a)(43)(G). Therefore, under the categorical approach we must determine whether the Washington robbery statute under which Alfred was convicted criminalizes conduct that falls outside the generic definition of theft.

We answered this question in Alvarado-Pineda. There, we held that "[a] comparison of the elements of the [Washington] statute to the elements of generic theft makes clear that the full range of conduct criminalized by Washington second-degree robbery falls within the meaning of generic theft." Alvarado-Pineda, 774 F.3d at 1203 (internal quotation marks omitted). We "conclude[d] that a conviction for Washington second-degree robbery, where accompanied by a sentence of at least one year, qualifies as an 'aggravated felony' under 8 U.S.C. § 1101(a)(43)(G)." Id. Neither Alfred nor the government has questioned the correctness of that decision. Rather, following our decision in Valdivia-Flores, the parties disagree over whether and how Washington's accomplice liability statute affects Alfred's robbery conviction. We thus turn to those questions.

IV. ACCOMPLICE LIABILITY

A. Whether Considering Accomplice Liability Is Necessary

1. Analysis

In Valdivia-Flores, we applied the categorical approach to Washington's aiding and abetting statute before looking at the statute of conviction. 876 F.3d at 1207. The government argues that doing so was incorrect. In taking up the question of accomplice liability in Valdivia-Flores, we relied on Gonzales v. Duenas-Alvarez, 549 U.S. 183, 127 S.Ct. 815, 166 L.Ed.2d 683 (2007). In Duenas-Alvarez, the Supreme Court reviewed our application of the categorical approach to Cal. Veh. Code § 10851, which criminalizes vehicle theft. 549 U.S. at 187, 127 S.Ct. 815. That statute punishes not only "[a]ny person who drives or takes a vehicle not his or her own," but also "any person who is a party or an accessory to or an accomplice in the driving or unauthorized taking or stealing." Cal. Veh. Code § 10851(a) (emphasis added). We had held that the statute's reference to accomplice liability made it overbroad, reasoning that aiding and abetting theft does not necessarily require a defendant to commit theft himself. Penuliar v. Ashcroft, 395 F.3d 1037, 1044-45 (9th Cir. 2005), rev'd by Duenas-Alvarez, 549 U.S. 183, 127 S.Ct. 815.

The Supreme Court disagreed. First, it observed that "all States and the Federal Government" had "expressly abrogated the distinction" between those "who actually committed the crime in question" and aiders and abettors. Duenas-Alvarez, 549 U.S. at 189-90, 127 S.Ct. 815 (internal quotation marks omitted). Thus, the Court concluded that generic theft encompasses the aiding and abetting of theft. Id. Next, the Court addressed whether California's accomplice liability scheme was "special," such that it would allow California to penalize "crime that falls outside the generic definition of 'theft.' " Id. at 190-94, 127 S.Ct. 815. The Court found no such distinction. Id. at 194, 127 S.Ct. 815.

Duenas-Alvarez left open the question of whether accomplice liability should be considered when a separate statute integrates accomplice liability into the statute of conviction. Such is the case here. Unlike the vehicle code at issue in Duenas-Alvarez, Washington's robbery statute does not expressly refer to aiding and abetting. See Wash. Rev. Code § 9A.56.190. Instead, a separate statute, Wash. Rev. Code § 9A.08.020(3), incorporates accomplice liability into all Washington crimes.

We hold that this is a distinction without a difference. The categorical approach is "[r]ooted in Congress' specification of conviction . . . as the trigger for immigration consequences." Mellouli, 575 U.S. at 806, 135 S.Ct. 1980. Though the statute of conviction is inextricably tied to the defendant's conviction, nothing confines the categorical analysis to a single statute. On the contrary, the Supreme Court has looked to statutes other than the statute of conviction to understand what a conviction may entail. See, e.g., Mathis, 579 U.S. at 507, 136 S.Ct. 2243; Taylor v. United States, 495 U.S. 575, 591-92, 110 S.Ct. 2143, 109 L.Ed.2d 607 (1990). Moreover, in defining the elements of a conviction, the Court often looks beyond the statute of conviction to see how state courts have applied the statute. See Mathis, 579 U.S. at 507, 136 S.Ct. 2243; Johnson v. United States, 559 U.S. 133, 138, 130 S.Ct. 1265, 176 L.Ed.2d 1 (2010); Duenas-Alvarez, 549 U.S. at 190-94, 127 S.Ct. 815. Although the statute of conviction is the center of our analysis, we cannot blind ourselves to sources that contribute to its meaning. We discern no relevant difference between (1) a statute that defines a crime and expressly punishes aiding and abetting and (2) a statute that defines a crime and is subject to another statute that expressly punishes aiding and abetting all crimes, including the particular crime in question.

A closer look at Washington law underscores why consideration of accomplice liability is important here. Washington, like the United States and all other states, has eliminated the distinction between principals and accomplices. By statute, in Washington, "[a] person is guilty of a crime if it is committed by the conduct of another person for which he or she is legally accountable." Wash. Rev. Code § 9A.08.020(1). Washington defines legal accountability broadly, encompassing all who are complicit in the commission of the crime. See Wash. Rev. Code § 9A.08.020(2)-(3); see also State v. Silva-Baltazar, 125 Wash.2d 472, 886 P.2d 138, 143 (1994) ("The complicity rule in Washington is that any person who participates in the commission of the crime is guilty of the crime and is charged as a principal."). Accordingly, Washington courts have declared that "[t]here is no separate crime of being an accomplice; [instead] accomplice liability is principal liability." State v. Handley, 115 Wash.2d 275, 796 P.2d 1266, 1276 (1990) (quoting State v. Toomey, 38 Wash.App. 831, 690 P.2d 1175, 1181 (1984)).

Washington's consolidation of accomplice and principal liability applies in all stages of prosecution. An information need not "expressly charge aiding or abetting or refer to other persons" for a defendant to be found guilty as an accomplice. State v. Rodriguez, 78 Wash.App. 769, 898 P.2d 871, 873 (1995); see also State v. Lynch, 93 Wash.App. 716, 970 P.2d 769, 772 (1999) ("[A]n information that charges an accused as a principal adequately apprises him of his potential liability as an accomplice."). Jurors may find a defendant guilty of a crime without agreeing whether the defendant acted as an accomplice or a principal. State v. Hoffman, 116 Wash.2d 51, 804 P.2d 577, 605 (1991) ("[I]t is not necessary that jurors be unanimous as to the manner of an accomplice's and principal's participation as long as all agree that they did participate in the crime."). And, "a defendant may be found guilty as a principal or as an accessory even though . . . the principal actor has not been tried or has been tried and acquitted." State v. Wilder, 25 Wash.App. 568, 608 P.2d 270, 274 (1980) (internal quotation marks and citations omitted); see also State v. Carothers, 84 Wash.2d 256, 525 P.2d 731, 734 (1974) ("[A] verdict may be sustained upon evidence that the defendant participated in the commission of the crime charged, as an aider or abetter, even though he was not expressly accused of aiding and abetting and even though he was the only person charged in the information."), abrogated on other grounds by State v. Harris, 102 Wash.2d 148, 685 P.2d 584, 587 (1984).

The upshot is that under Washington law it is impossible to determine whether an individual defendant has been convicted as a principal or an accomplice—or even whether a defendant acted alone, in concert, or with the aid of another—without looking to the facts underlying the conviction. Yet the categorical approach forbids us from doing so. See Descamps, 570 U.S. at 261, 133 S.Ct. 2276. Because a conviction under Wash. Rev. Code § 9A.56.190 does not necessarily establish that Alfred acted as a principal, we must consider the possibility he acted as an accomplice. We therefore conclude that we must consider Washington accomplice liability in our categorical analysis of Washington robbery. See Bourtzakis v. U.S. Att'y Gen., 940 F.3d 616, 621-22 (11th Cir. 2019) (arriving at the same conclusion).

2. Response to the Concurring Opinion

Our dissenting colleagues agree with us on this point. Dissenting Op. at 1067-68. Our concurring colleagues, however, disagree with this conclusion. Although the concurrence believes that Washington aiding and abetting law is too different from federal generic aiding and abetting to be a categorical match, see Concurring Op. at 1035-66, it nevertheless concludes that our categorical analysis should not address accomplice liability at all, id. at 1058. According to the concurrence, this is because "the issue of aiding and abetting liability is not subsumed into one of the constituent elements of the underlying predicate offense," but rather is "a theory of liability" that does not affect what a jury must find to convict a defendant. Id. at 1053.

We agree with many of the building blocks of the concurrence's analysis. We agree, for example, that the categorical approach compares the elements of the crime in question with the elements of the generic crime. Concurring Op. at 1052. As the concurrence explains, and we agree, an element may be comprised of alternative means or methods for proving a particular element, but "that overarching element subsumes those alternative means." Id. We further agree "that aiding and abetting is merely a means of finding principal liability, that it does not itself constitute an element of the underlying offense, and that jurors need not unanimously agree as to whether a defendant is guilty as a principal or as an aider and abettor." Id. at 1052.

Where we disagree with the concurrence is its conclusion that "aiding and abetting liability is not subsumed into one of the constituent elements of the underlying predicate offense" and that, therefore, we must " 'disregard[ ]' aiding and abetting liability." Id. at 1053 (emphasis added) (quoting Mathis, 579 U.S. at 517, 136 S.Ct. 2243). That conclusion does not follow from the concurrence's premises because it mischaracterizes what abrogation of the distinction between principals and accomplices means.

In general, under modern accomplice liability theory, "aiding and abetting is a different means of committing a single crime," it is "not a separate offense itself," and "the government ha[s] no obligation to elect between charging a substantive offense and charging liability on an aiding and abetting theory." United States v. Garcia, 400 F.3d 816, 820 (9th Cir. 2005). Accordingly, "aiding and abetting is embedded in every federal indictment for a substantive crime." Id. Although it is still useful for us to refer to a "principal" or an "accomplice," in the end they are equally culpable and may be convicted of the same offense. We nevertheless continue to traverse the metaphysical line between the two as a way of distinguishing degrees of culpability, which properly plays a role in sentencing. See United States v. Smith, 891 F.2d 703, 714 (9th Cir. 1989) ("The evidence was sufficient to make him an aider and abettor . . . . The mildness of his sentence corresponds to the minor role he had."); see also United States v. Dominguez-Caicedo, 40 F.4th 938, 959-62 (9th Cir. 2022); United States v. Cordova Barajas, 360 F.3d 1037, 1042-43 (9th Cir. 2004); U.S.S.G. §§ 2X2.1 cmt. n.1 ("aiding and abetting the commission of an offense has the same offense level as the underlying offense. An adjustment for a mitigating role (§ 3B.2) may be applicable"), 3B1.2 & cmts. nn.3-5 (describing participants who are "less culpable," "among the least culpable," and "substantially less culpable.").

The concurrence's misstep results from mischaracterizing Mathis. In Mathis, the Supreme Court took up the question of how to apply the categorical approach "when a statute happens to list various means by which a defendant can satisfy an element." 579 U.S. at 508, 136 S.Ct. 2243. The ACCA imposes a mandatory minimum sentence for defendants who have three prior convictions for a "violent felony." 18 U.S.C. § 924(c). As relevant to Mathis, the ACCA defines "violent felony" as any felony that is "burglary, arson, or extortion." Id. § 924(c)(2)(B)(ii). The statute at issue was Iowa's burglary statute, which, through another statute, allowed conviction for burglary for unlawful entry into "any building, structure, [or] land, water, or air vehicle." Mathis, 579 U.S. at 507, 136 S.Ct. 2243 (quoting Iowa Code § 702.12). The parties and the Court agreed that no other state permitted a burglary conviction for entering a "vehicle." Id. at 512, 136 S.Ct. 2243. Thus, "some but not all" the locations listed in the Iowa burglary statute "satisf[ied] the generic definition of burglary." These different locations were not alternative elements—creating separate crimes—but rather "alternative ways of satisfying a single locational element." Id. at 507, 512, 136 S.Ct. 2243. But because an Iowa jury "need not agree whether the burgled location was a building, other structure, or vehicle," a burglary conviction in Iowa did not necessarily mean that a defendant satisfied the locational element for burglary in a way that fit the generic definition. Id. at 517-18, 136 S.Ct. 2243. As a result, the Court found Iowa's burglary statute to be categorically overbroad and that held Mathis's burglary conviction could not be used to enhance his sentence under the ACCA. Id. at 515-17, 520, 136 S.Ct. 2243.

The concurrence reads Mathis to mean that, because an alternative means of satisfying an element is not itself an element, a state's aiding and abetting scheme has no bearing on our categorical analysis. See Concurring Op. at 1051-54. We think, to the contrary, because accomplice liability is one means of satisfying an element necessary to support a conviction, we must consider it. If, as Alfred claims, accomplice liability requires a lesser mens rea than principal liability, then the choice between those alternative means is significant. A simple example will prove our point. Suppose Washington law provided that "The state may prove a theft offense through evidence that the defendant acted purposely or negligently." See Model Penal Code § 2.02(2)(a), (d) (describing those mentes reae). In the same way the Iowa burglary statute at issue in Mathis provided alternate means for satisfying the locational element, this statute provides alternative means for satisfying the mens rea element—purpose and negligence. Under our hypothetical Washington statute, a defendant may be convicted of a theft offense if he acted with either mens rea, and a Washington jury would not have to decide unanimously whether the defendant acted purposefully or negligently. In such circumstances, we would presume that "the state conviction 'rested upon the least of the acts' criminalized by the statute." Esquivel-Quintana, 581 U.S. at 389, 137 S.Ct. 1562 (cleaned up) (quoting Johnson, 559 U.S. at 137, 130 S.Ct. 1265)). But permitting conviction on the basis of negligence would not match the elements of generic theft. See infra pp. 1035-41; cf. United States v. Benally, 843 F.3d 350, 354 (9th Cir. 2016) (holding that a conviction based on a mens rea of gross negligence is not a "crime of violence" under 18 U.S.C. § 924(c)); cf. also Borden, 141 S. Ct. at 1825-28 (concluding that reckless conduct does not qualify as a "violent felony" under the ACCA); United States v. Garcia-Jimenez, 807 F.3d 1079, 1085 (9th Cir. 2015) (concluding that "a mens rea of extreme indifferent recklessness is not sufficient to meet the federal generic definition of aggravated assault"). In our hypothetical, negligence (as an alternative mens rea) is subsumed in the intent element, and no one could reasonably argue that "we must 'disregard[ ]' [the mens rea] in determining whether there is a categorical match between a state offense and a federal generic offense." Concurring Op. at 1053 (quoting Mathis, 579 U.S. at 517, 136 S.Ct. 2243).

The concurrence claims that our hypothetical "is not relevant" because, in its view, aiding and abetting is different from a crime that lists two possible mentes reae. Concurring Op. at 1055. According to the concurrence, this is because—unlike an offense that lists multiple means of commission within the statute—"[w]hen aiding and abetting is invoked . . . the jury must still find that, as to either or both of the co-participants, each of the elements of that offense have been met." Id. at 1053 (emphasis added). This statement is technically correct, but it only serves to undercut the concurrence's position.

We agree with the concurrence that, to convict a defendant as an accomplice, "the Government . . . 'must prove that someone committed,' " Concurring Op. at 1053 (emphasis added) (quoting United States v. Mann, 811 F.2d 495, 497 (9th Cir. 1987)), "all the acts constituting the elements of the substantive crime charged," Mann, 811 F.2d at 497. However, the concurrence also appears to believe that it does not matter whether a conviction necessarily establishes that the defendant in question committed a generic offense so long as the conviction establishes that someone, somewhere did so. See Concurring Op. at 1054-55. Under the concurrence's theory, a defendant could receive an enhanced sentence because someone else committed a generic offense.

The concurrence's position makes no sense. A defendant—even an accomplice—is responsible for his own acts, not the acts of others. See Mann, 811 F.2d at 497 ("It is not a prerequisite to a conviction for aiding and abetting . . . that the principal be tried and convicted, or even that the principal be identified. In fact, an aider and abettor's conviction may be upheld even though the principal is acquitted of the underlying offense." (citations omitted)). That is why the categorical approach asks "whether a state offense necessarily involves the defendant's" commission of acts that categorically match a federal standard. Borden v. United States, — U.S. —, 141 S. Ct. 1817, 1822, 210 L.Ed.2d 63 (2021) (plurality op.) (emphasis added). A conviction showing that someone else satisfied the elements of a generic offense does not necessarily mean that the defendant satisfied those same elements. By ignoring this basic principle, the concurrence would tie immigration and sentencing consequences to facts that a conviction does not necessarily establish, thereby violating basic principles of the categorical approach. See Moncrieffe, 569 U.S. at 205 n.11, 133 S.Ct. 1678 (the categorical approach focuses on "what facts are necessarily established by a conviction for the state offense"); see also Mathis, 579 U.S. at 505, 136 S.Ct. 2243.

In sum, we remain perplexed that the concurrence concludes that Washington accomplice liability sweeps more broadly than generic law, see Concurring Op. at 1059-66, but simultaneously insists that accomplice liability is "not relevant" to our categorical inquiry, id. at 1055. To be sure, we disagree with the concurrence's conclusion that Washington accomplice liability is not a categorical match for generic aiding and abetting. But the concurrence would have us ignore any overbreadth that accompanies a state's accomplice liability scheme, no matter how egregious.

***

We think the question raised in Valdivia-Flores—whether a defendant can be convicted in Washington of aiding and abetting based on a lesser mens rea than that required for principal liability—is not irrelevant and must be answered. We now turn to that question. B. Defining Washington and Generic Accomplice Liability

1. Washington accomplice liability

Washington law defines a person as an accomplice if:

(a) With knowledge that it will promote or facilitate the commission of the crime, he or she:

(i) Solicits, commands, encourages, or requests such other person to commit it; or

(ii) Aids or agrees to aid such other person in planning or committing it; or

(b) His or her conduct is expressly declared by law to establish his or her complicity.

Washington courts have held that the knowledge component in the accomplice liability statute refers to knowledge of the conduct that forms the basis of the charged offense. State v. Roberts, 142 Wash.2d 471, 14 P.3d 713, 735 (2000). Thus, to convict a defendant as an accomplice under Washington law, the government must show that the defendant "acted with knowledge that his or her conduct would promote or facilitate the crime" charged, rather than knowledge that the conduct would "aid[ ] in the commission of any crime." State v. Cronin, 142 Wash.2d 568, 14 P.3d 752, 757 (2000). This means that accomplice liability "[does] not extend beyond the crimes of which the accomplice actually has knowledge." Id. at 758 (quoting Roberts, 14 P.3d at 735). At the same time, "[s]pecific knowledge of the elements of the coparticipant's crime need not be proved to convict one as an accomplice." State v. Dreewes, 192 Wash.2d 812, 432 P.3d 795, 801 (2019) (quoting State v. Rice, 102 Wash.2d 120, 683 P.2d 199 (1984)). Instead, the government need only prove the defendant had "general knowledge of his coparticipant's substantive crime." Roberts, 14 P.3d at 736 (quoting Rice, 683 P.2d at 203). In practical terms, this means that the government must establish the accomplice knew of the general crime (i.e., "assault"), but it need not establish knowledge of the specific degree of the crime (i.e., "first-degree assault"). Sarausad v. State, 109 Wash.App. 824, 39 P.3d 308, 314-15 (2001) (discussing Cronin, 14 P.3d at 758-59).

2. Generic accomplice liability

To determine the federal generic definition of a crime, we "ordinarily survey[ ] a number of sources—including state statutes, the Model Penal Code, federal law, and criminal treatises." United States v. Garcia-Jimenez, 807 F.3d 1079, 1084 (9th Cir. 2015). "Most often, '[t]he generic definition of an offense roughly corresponds to the definitions of the offense in a majority of the States' criminal codes,' " id. (quoting United States v. Garcia-Santana, 774 F.3d 528, 534 (9th Cir. 2014), abrogated on other grounds by Esquivel-Quintana, 581 U.S. 385, 137 S.Ct. 1562), but how courts have applied these statutes is also relevant, see Duenas-Alvarez, 549 U.S. at 190-91, 127 S.Ct. 815. The generic offense is simply "the offense as commonly understood." Mathis, 579 U.S. at 503, 136 S.Ct. 2243.

Commentators have noted that "[c]onsiderable confusion exists as to what the accomplice's mental state must be in order to hold him accountable for an offense committed by another." 2 Wayne R. LaFave, Substantive Criminal Law § 13.2(b), at 466 (3d ed. 2018) [hereinafter LaFave]. The confusion has resulted in "considerable variation in the language used by courts and legislatures" regarding the mens rea necessary for accomplice liability. Id.; see also Jens David Ohlin, Wharton's Criminal Law § 10.9 (16th ed. 2022) [hereinafter Wharton's Criminal Law] ("How the general complicity mental standard applies to specific crimes has generated a rich but sometimes confusing jurisprudence . . . ."); Alexander F. Sarch, Condoning the Crime: The Elusive Mens Rea for Complicity, 47 Loy. U. Chi. L.J. 131, 133 (2015) (observing "a long history of disagreement about how, precisely, this mens rea should be understood" and cataloguing various approaches including "intention or purpose" and "knowledge"); Baruch Weiss, What Were They Thinking: The Mental States of the Aider and Abettor and the Causer Under Federal Law, 70 Fordham L. Rev. 1341, 1351 (2002) ("[T]he current status of the law on the aider and abettor's mental state is far from clear. In fact, it is best described today as in a state of chaos.").

We may divide state accomplice liability law into four general categories. See John F. Decker, The Mental State Requirement for Accomplice Liability in American Criminal Law, 60 S. C. L. Rev. 237 (2008) (surveying federal and state law). These categories are based in both state statutory language as well as state common law.

Purpose. In the first category are jurisdictions that require a mens rea of "purpose" or similar term like "willfulness." Federal statutory definitions, the Model Penal Code, and some states fall into this category. See 18 U.S.C. § 2(b) ("Whoever willfully causes an act to be done which if directly performed by him or another would be an offense against the United States, is punishable as a principal." (emphasis added)); Model Penal Code § 2.06(3)(a) ("A person is an accomplice of another person . . . if[,] with the purpose of promoting or facilitating the commission of the offense, he . . . aids . . . such other person in planning or committing it." (emphasis added)); N.J. Stat. Ann. § 2C:2-6(c) (using Model Penal Code language). A number of other states use the broader term "intent." See, e.g., Or. Rev. Stat. § 161.155 ("A person is criminally liable for the conduct of another person constituting a crime if . . . [w]ith the intent to promote or facilitate the crime the person . . . [a]ids or abets . . . such other person in planning or committing the crime." (emphasis added)); Tex. Penal Code § 7.02(a) ("A person is criminally responsible for an offense committed by the conduct of another if . . . acting with intent to promote or assist the commission of the offense, he solicits, encourages, directs, aids, or attempts to aid the other person to commit the offense[.]" (emphasis added)). Roughly half the states use one of these terms. LaFave, supra, § 13.2(b), at 466-67 nn.70 & 72 (listing states).

Knowledge. In the second category are states like Washington that require a mens rea of knowledge. See Wash. Rev. Code § 9A.08.020(3); Ind. Code Ann. § 35-41-2-4 ("A person who knowingly or intentionally aids, induces, or causes another person to commit an offense commits that offense . . . ." (emphasis added)); W. Va. Code Ann. § 61-2-14e ("If any person in any way knowingly aid or abet any other person in the commission of any offense . . . such person so aiding and abetting shall be guilty as a principal . . . ." (emphasis added)); Wyo. Stat. Ann. § 6-1-201 ("A person who knowingly aids or abets in the commission of a felony . . . may be indicted, informed against, tried and convicted as if he were a principal . . . ." (emphasis added)). Few states use statutory language that requires only knowledge. See Decker, supra, at 250-51 (placing four states in this category).

Derivative Mens Rea. Third, roughly fifteen states require the accomplice to act with the same mens rea as that required to prove the principal offense. Decker, supra, at 275-310 (surveying states); see, e.g., N.Y. Penal Law § 20.00 ("When one person engages in conduct which constitutes an offense, another person is criminally liable for such conduct when, acting with the mental culpability required for the commission thereof, he . . . intentionally aids such person to engage in such conduct." (emphasis added)); Utah Code Ann. § 76-2-202 ("Every person, acting with the mental state required for the commission of an offense who . . . intentionally aids another person to engage in conduct which constitutes an offense shall be criminally liable as a party for such conduct." (emphasis added)). This has been called the "derivative approach." Sarch, supra, at 133. If a principal may be convicted for a crime based on criminally negligent conduct, so too may an accomplice. Weiss, supra, at 1376. At the same time, evidence of the accomplice's purposeful intent is necessary to support a conviction for a crime that requires the principal to act purposefully. Id.

Natural and Probable Consequences. The fourth category comprises states that have expanded accomplice liability through use of the natural and probable consequences doctrine, derived from tort law. See People v. Prettyman, 14 Cal.4th 248, 58 Cal.Rptr.2d 827, 926 P.2d 1013, 1037-38 (1996) (Brown, J., concurring in the judgment) (summarizing history). This doctrine "holds a person accountable not just for the crimes the person intended to aid and abet but also for any offense that is a reasonably foreseeable consequence of the criminal scheme." Decker, supra, at 249; see also Ariz. Rev. Stat. § 13-303(A)(3) ("The person is an accomplice of such other person in the commission of an offense including any offense that is a natural and probable or reasonably foreseeable consequence of the offense for which the person was an accomplice."). A specific mens rea is still necessary for the initial offense, but not for the offenses that foreseeably follow. See State v. Linscott, 520 A.2d 1067, 1070 (Me. 1987) ("So long as the accomplice intended to promote the primary crime, and the commission of the secondary crime was a foreseeable consequence of the accomplice's participation in the primary crime, no further evidence of the accomplice's subjective state of mind as to the secondary crime is required."). As a result, this category overlaps with the other three; states in this category require a purposeful, knowing, or derivative mens rea for the initial crime, but make the accomplice liable for foreseeable crimes committed by the principal. See, e.g., Minn. Stat. Ann. § 609.05(1)-(2) (purposeful); Iowa Code § 703.2 (knowing); Ariz. Rev. Stat. § 13-303(A)(2)-(3) (derivative). Over a dozen states use some variation of this rule. See LaFave, supra, § 13.3(b), at 494-95 nn.28-29; Decker, supra, at 312-52.

Although these categories embrace different language, it is difficult to tell whether the categories actually are inconsistent with each other or simply represent different approaches to defining a single concept of accomplice liability. See LaFave, supra, § 13.2(b), at 467 (suggesting that the variations in language "may represent different ways of stating the same mental state requirement"). From this, it is equally difficult to discern a clear generic standard. However, we can say that a majority of jurisdictions use statutory language that requires a showing of purpose or something similar to convict a defendant as an accomplice. This majority includes all the states in the first category, and, at least for specific intent crimes, the states in the third category. States in the fourth category are included inasmuch as they incorporate an intentional or a derivative mens rea requirement for the initial offense.

A majority of state statutes require some kind of purposeful conduct to establish accomplice liability for at least some crimes, but jurisdictions have defined that intent differently. See Wharton's Criminal Law, supra, at § 10:9 ("The intent requirement for accomplice liability is sometimes ambiguous in the sense that intent can mean 'purposely' and at other times 'knowingly.' "). Thus, to define generic accomplice liability, we must look at how courts have applied statutory intent requirements. In the context of this case, our task is to determine the relationship between jurisdictions that require proof of "purpose," "willfulness," or "intent," and those that, like Washington, require proof of "knowledge." We find that, in this context, these terms are largely synonymous.

The question has some history in the Supreme Court. In United States v. Bailey, 444 U.S. 394, 100 S.Ct. 624, 62 L.Ed.2d 575 (1980), the Court took up a general discussion of mens rea. The Court began with the declaration that "[f]ew areas of criminal law pose more difficulty than the proper definition of the mens rea required for any particular crime." Id. at 403, 100 S.Ct. 624. The Court observed that the Model Penal Code had sought to replace "the ambiguous and elastic term 'intent' . . . with a hierarchy of culpable states of mind." Id. at 404, 100 S.Ct. 624. "[I]n descending order of culpability," they are "purpose, knowledge, recklessness, and negligence." Id. (footnote omitted). The Court described the first two, "purpose" and "knowledge," as "the most significant, and most esoteric," and said that "[i]n the case of most crimes, 'the limited distinction between knowledge and purpose has not been considered important.' " Id. (quoting United States v. United States Gypsum Co., 438 U.S. 422, 445, 98 S.Ct. 2864, 57 L.Ed.2d 854 (1978)). But, the Court acknowledged, "[i]n certain narrow classes of crimes," including inchoate crimes, "culpability has been thought to merit special attention." Id. at 405, 100 S.Ct. 624; see also Borden, 141 S. Ct. at 1822-23 & n.3 (listing "accessory liability (aiding and abetting)" as among the crimes, including inchoate crimes, requiring "special attention"; expressing no view on "those offenses, nor the relationship more generally between purpose and knowledge" (internal quotation marks and citations omitted)). The Court offered this distinction:

A person who causes a particular result is said to act purposefully if he consciously desires that result, whatever the likelihood of that result happening from his conduct, while he is said to act knowingly if he is aware that the result is practically certain to follow from his conduct, whatever his desire may be as to that result.

If Bailey's general observation were the last word, we would be compelled to agree that Washington's requirement of "knowledge" is a lesser mens rea than "intent" or "purpose." See Dissenting Op. at 1067-69. But if the "ambigu[ity]" and "elastic[ity]" that inheres in "the limited distinction between knowledge and purpose" has not acquired clarity in the forty years since the Court identified the conundrum in Bailey, 444 U.S. at 404, 100 S.Ct. 624 (internal quotation marks omitted), we at least have some additional clarity with respect to accomplice liability.

We find the Court's general discussion of accomplice liability in Rosemond v. United States, 572 U.S. 65, 134 S.Ct. 1240, 188 L.Ed.2d 248 (2014), helpful. Rosemond involved the federal aiding and abetting statute, 18 U.S.C. § 2, which provides that a person may be "punishable as a principal" if he "willfully causes an act to be done which if directly performed by him or another would be an offense against the United States." Id. § 2(b). The Court observed that § 2 is derived from common-law standards that punish a person for aiding or abetting if he took "an affirmative act in furtherance of that offense . . . with the intent of facilitating the offense's commission." Rosemond, 572 U.S. at 71, 134 S.Ct. 1240. Putting that intent in practical terms, the Court stated that the "intent requirement [is] satisfied when a person actively participates in a criminal venture with full knowledge of the circumstances constituting the charged offense." Id. at 77, 134 S.Ct. 1240; see Central Bank of Denver, N.A. v. First Interstate Bank of Denver, N.A., 511 U.S. 164, 181, 114 S.Ct. 1439, 128 L.Ed.2d 119 (1994) ("[T]hose who provide knowing aid to persons committing federal crimes, with the intent to facilitate the crime, are themselves committing a crime."); U.S. Gypsum Co., 438 U.S. at 445, 98 S.Ct. 2864 (recognizing that, regardless of whether a defendant acted purposefully or knowingly, it may be said that she intended the result of her actions); Pereira v. United States, 347 U.S. 1, 12, 74 S.Ct. 358, 98 L.Ed. 435 (1954) (holding that "knowledge" of the scheme satisfied the intent requirement for accomplice liability). Throughout its opinion, the Court equated intent with "advance knowledge," meaning "knowledge at a time the accomplice can do something with it—most notably, opt to walk away." Rosemond, 572 U.S. at 78, 134 S.Ct. 1240 (footnote omitted). Accordingly, "[w]hat matters for purposes of gauging intent . . . is that the defendant has chosen, with full knowledge, to participate in the illegal scheme—not that, if all had been left to him, he would have planned the identical crime." Id. at 79, 134 S.Ct. 1240. In dissent, Justice Alito observed:

[S]ome of our cases suggest that an aider and abettor must act purposefully or with intent. . . . [Other] cases . . . appear to hold that the requisite mens rea is simply knowledge. The Court refers interchangeably to both of these tests and thus leaves our case law in the same, somewhat conflicted state that previously existed. But because the difference between acting purposefully . . . and acting knowingly is slight, this is not a matter of great concern.

The blurred line between intent and knowledge is apparent throughout state law as well. Many states with accomplice liability statutes that require a showing of intent nevertheless allow a conviction based on circumstantial evidence of the putative accomplice's knowledge of the crime. See generally LaFave, supra § 13.2(b), at 468 (noting authority "to the effect that one may become an accomplice by giving encouragement or assistance with knowledge that it will promote or facilitate a crime"). For example, Missouri defines an accomplice as one who "aids or agrees to aid" another "with the purpose of promoting the commission of an offense." Mo. Rev. Stat. § 562.041. Nevertheless, a conviction may be supported by "[a]ny evidence, either direct or circumstantial, that shows affirmative participation," including the accomplice's conduct and knowledge of the offense. State v. Barker, 442 S.W.3d 165, 169 (Mo. App. W.D. 2014) (internal quotation marks omitted); see also State v. Williams, 409 S.W.3d 460, 468 (Mo. App. W.D. 2013) ("Circumstances that may support the inference of an accomplice's affirmative participation include . . . knowledge." (internal quotation marks omitted)). Likewise, in Michigan, accomplice liability attaches if the defendant "intended the commission of the crime or had knowledge that the principal intended [the crime's] commission at the time that [the defendant] gave aid and encouragement." People v. Robinson, 475 Mich. 1, 715 N.W.2d 44, 48 (2006) (internal quotation marks omitted and emphasis added); see also Mich. Comp. Laws § 767.39 (attaching criminal liability to "[e]very person concerned in the commission of an offense," without providing a mens rea). And Wisconsin, which requires that an accomplice act "intentionally," Wis. Stat. Ann. § 939.05(2)(b), defines that intent in terms of knowledge. See Wis. JI-Criminal, No. 405 ("A person intentionally aids or abets the commission of a crime when, acting with knowledge or belief that another person is committing or intends to commit a crime, he knowingly . . . renders aid to the person who commits the crime."); see also State v. Hibbard, 404 Wis.2d 668, 982 N.W.2d 105, 110 (App. 2022) (approving the instruction). These states are not outliers; many others have mingled knowledge and intent in the accomplice liability context.

We hold that generic accomplice liability requires a showing that the putative accomplice intentionally aided or abetted another in the commission of the crime. However, because many jurisdictions have interpreted this intent requirement as encompassing both purposeful and knowing conduct, we also conclude that advance knowledge of the crime is sufficient to support a conviction for generic accomplice liability.

This holding does not mean that all statutes that fall under the generic definition of accomplice liability are to be interpreted in the same way. Doing so would ignore the jurisdictional distinctions that we have so carefully described. See supra pp. 1038-39. Instead, we observe that aiding and abetting has been defined in so many different ways that its generic form must be broad enough to encompass jurisdictions that require purposeful conduct, as well as those that require accomplices to act with knowledge that their conduct will promote or facilitate commission of the crime. See Wash. Rev. Code § 9A.08.020(3).

Our broad interpretation of a generic accomplice liability does not erase the distinctions between states' approaches to accomplice liability. Duenas-Alvarez illustrates this point. There, the Court held that California's natural and probable consequences doctrine adequately reflected generic law so as not to be overbroad. See Duenas-Alvarez, 549 U.S. at 193-94, 127 S.Ct. 815. But this holding did not impose the natural and probable consequences doctrine on the states that did not already use it. Nor did it render every accomplice liability statute that did not use a natural and probable consequences doctrine overbroad. Rather, Duenas-Alvarez envisions a generic definition of accomplice liability that encompasses jurisdictions that use a natural and probable consequences doctrine, as well as those that do not. See id. at 190-91, 193-94, 127 S.Ct. 815. C. Comparing Washington and Generic Accomplice Liability

Having concluded that the intent element of generic accomplice liability may be satisfied by showing advance knowledge of the crime, it remains for us to determine whether Washington's mens rea requirement diverges significantly from this standard. There are at least two ways to compare Washington law to the generic standard. First, we can simply compare the text of Washington's aiding and abetting standard to the generic standard to see if "greater breadth is evident from its text." Lopez-Aguilar v. Barr, 948 F.3d 1143, 1147 (9th Cir. 2020) (quoting United States v. Grisel, 488 F.3d 844, 850 (9th Cir. 2007) (en banc), abrogated on other grounds by United States v. Stitt, — U.S. —, 139 S. Ct. 399, 202 L.Ed.2d 364 (2018)). Second, we can review Washington's case law to see whether there is a "a realistic probability, not a theoretical possibility, that the State would apply its statute to conduct that falls outside the generic definition of a crime." Duenas-Alvarez, 549 U.S. at 193, 127 S.Ct. 815; see also United States v. Taylor, — U.S. —, 142 S. Ct. 2015, 2024-25, 213 L.Ed.2d 349 (2022) (interpreting the "realistic probability" requirement narrowly). In this case, we will consider both because we do not think either approach, considered alone, fully captures Washington's state of the law. See Alvarado-Pineda, 774 F.3d at 1202 (looking to Washington statutes and cases).

We have been aided in our inquiry by the Eleventh Circuit's analysis of this same question in Bourtzakis v. United States Attorney General, 940 F.3d 616 (11th Cir. 2019). In that case, the Eleventh Circuit had to determine whether a conviction of drug trafficking under Wash. Rev. Code § 69.50.401(a) constitutes an aggravated felony for purposes of the INA. Id. at 618. After determining that Washington's accomplice liability scheme required it to take up the question of generic accomplice liability, the Eleventh Circuit looked to whether Washington aiding and abetting was broader than federal law. Although the Eleventh Circuit acknowledged a "difference in language," it nevertheless found that "the mens-rea requirements for accomplice liability under the Washington statute and the federal Act do not diverge." Id. at 622.

Reviewing Supreme Court precedent, including Rosemond, the Eleventh Circuit determined that "the requirement that an accomplice to a federal crime 'intend[ ] to facilitate that offense's commission is satisfied by proof that the accomplice actively participated in the crime and knew the nature of the crime he was facilitating.' " Id. at 623 (quoting Rosemond, 572 U.S. at 76-77, 134 S.Ct. 1240 (majority op.)). Turning to Washington precedent, the Eleventh Circuit discussed Washington's rule that, " 'to be an accomplice, an individual must have acted with knowledge that he or she was promoting or facilitating the crime for which that individual was eventually charged'—not just 'any crime.' " Id. (quoting Cronin, 14 P.3d at 757-58) (cleaned up). Comparing the two standards, the Eleventh Circuit concluded that "the mens-rea requirements for accomplice liability under Washington and federal law mirror one another" because, under both, "a person is liable as an accomplice if he actively participates in a crime and knows the nature of the crime he is facilitating." Id. at 623.

We agree with the Eleventh Circuit's reasoning in Bourtzakis. Both Washington and federal generic accomplice liability law require prosecutors to meet the same standard of proof: they must show the accomplice had advance knowledge of the crime he facilitated. In Washington, the statute provides that an accomplice acts "[w]ith knowledge that [the conduct] will promote or facilitate the commission of the crime." Wash. Rev. Code § 9A.08.020(3)(a). Washington courts have reaffirmed the statutory requirement, holding that the state must prove a defendant "acted with knowledge that he or she was promoting or facilitating the crime for which that individual was eventually charged." Cronin, 14 P.3d at 758. And the Supreme Court has interpreted common law accomplice liability as requiring evidence that a defendant "actively participat[ed] in a criminal venture with full knowledge of the circumstances constituting the charged offense." Rosemond, 572 U.S. at 77, 134 S.Ct. 1240. We see no significant daylight between these two standards.

Citing Rosemond, Alfred posits that the standards are distinct because generic federal law requires the accomplice have "full awareness," id. at 76-77, 134 S.Ct. 1240, whereas Washington law requires only "general knowledge," Roberts, 14 P.3d at 736. The dissent advances a similar argument, contrasting Rosemond's requirement that the accomplice be aware of the "entire crime" with Washington law providing that the state need not prove an accomplice was aware of "every element" of the crime. Dissenting Op. at 1069-70 (first quoting Rosemond, 572 U.S. at 76, 134 S.Ct. 1240, then quoting Roberts, 14 P.3d at 736). To answer these arguments, we must first explain the facts of Rosemond in greater detail.

Rosemond tasked the Supreme Court with determining "what the Government must show when it accuses a defendant of aiding or abetting" a violation of 18 U.S.C. § 924(c). 572 U.S. at 67, 134 S.Ct. 1240. Section 924(c) prohibits "us[ing] or carr[ying] a firearm during and in relation to any crime of violence or drug trafficking crime." It is an unusual statute because it penalizes the use of a firearm during the commission of another crime. Section 924(c) defines a "combination crime" that "punishes the temporal and relational conjunction of two separate acts"—use of a firearm and commission of a violent or drug trafficking offense—"on the ground that together they pose an extreme risk of harm." Rosemond, 572 U.S. at 75, 134 S.Ct. 1240. Rosemond was convicted of violating § 924(c), but he could also have been charged with aiding and abetting a marijuana sale even if a gun had not been used. The Court held that, to be convicted under § 924(c) as an accomplice, Rosemond must have "advance knowledge" of the firearm, and not just the marijuana deal, such that he may "make the relevant legal (and indeed moral) choice." Id. at 78, 134 S.Ct. 1240.

Alfred and the dissent contrast the Court's holding in Rosemond with several Washington cases, the most persuasive of which is State v. Davis, 101 Wash.2d 654, 682 P.2d 883 (1984). There, the Washington Supreme Court held that, to convict an individual of first-degree armed robbery with a deadly weapon, the prosecution was "not required to prove that the accomplice had knowledge that the principal was armed." Id. at 884; see also State v. McChristian, 158 Wash.App. 392, 241 P.3d 468, 472 (2010) (holding that, to be convicted of first-degree assault with a deadly weapon, an accomplice need not have knowledge that the principal would use a deadly weapon). Alfred argues that this runs contrary to Rosemond, which requires such prior knowledge for a § 924(c) conviction. Alfred's argument fails for three independent reasons.

First, neither Davis nor any other Washington case he cites involved a combination crime like § 924(c). Section 924(c) is a "freestanding offense" that requires proof of two distinct acts. Rosemond, 572 U.S. at 75, 134 S.Ct. 1240. In contrast, first-degree armed robbery is an enhanced version of a base offense—simple robbery. See Wash. Rev. Code §§ 9A.56.190, 9A.56.200. We do not know what Washington courts would do if presented with a combination crime like § 924(c). However, Washington precedent indicates Washington courts would likely require the same double-barreled knowledge standard that Rosemond applied to § 924(c). Washington law provides that an accomplice must act with "the purpose to promote or facilitate the particular conduct that forms the basis for the charge." Sarausad, 39 P.3d at 314 (internal quotation marks and emphasis omitted). A combination crime "contain[s] two distinct conduct elements." United States v. Rodriguez-Moreno, 526 U.S. 275, 280, 119 S.Ct. 1239, 143 L.Ed.2d 388 (1999); see also Dean v. United States, 556 U.S. 568, 576, 129 S.Ct. 1849, 173 L.Ed.2d 785 (2009) (referring to § 924(c) as punishing "unlawful conduct twice over"). Thus, under Washington law an accomplice cannot act with the purpose to promote a combination crime without having knowledge of both distinct conduct elements—the same standard used in Rosemond. At the same time, nothing in Davis suggests that Washington courts would interpret combination crimes in the same way they interpret enhanced offenses. Accordingly, Davis does not stand for the proposition that Washington accomplice liability extends beyond that expressed in Rosemond.

Second, Rosemond's discussion of § 924(c) does not purport to define generic accomplice liability law. Rather, it reflects the challenge of applying federal accomplice liability law to a combination crime. Though much of Rosemond builds on general principles, see id. at 70-71, 76-77, 134 S.Ct. 1240, its application of those principles to § 924(c) is novel. Consequently, some of our sister circuits have declined to apply Rosemond's two-headed advance knowledge requirement outside the context of § 924(c). See, e.g., United States v. Simpson, 44 F.4th 1093, 1098-99 (8th Cir. 2022); United States v. Chavez, 951 F.3d 349, 361-62 (6th Cir. 2020); United States v. Baker, 912 F.3d 297, 313-14 (5th Cir. 2019); Stuckey v. United States, 878 F.3d 62, 71-72 (2d Cir. 2017). And, more importantly, many states have explicitly rejected Rosemond's interpretation of § 924(c) as incompatible with state law. See, e.g., Hicks v. State, 295 Ga. 268, 759 S.E.2d 509, 514-15 & n.3 (2014); People v. Sandoval, 488 P.3d 441, 445-46 (Colo. Ct. App. 2018); State v. Ward, 473 S.W.3d 686, 693 (Mo. Ct. App. 2015); see also State v. Edwards, No. 45764-4-II, 2016 WL 900668, at *9 n.12 (Wash. Ct. App. Mar. 1, 2016). Contrary to the concurrence's assertions otherwise, see Concurring Op. at 1065-66, we express no opinion on whether the "full knowledge" requirement in Rosemond applies to crimes that are not combination crimes. Rather, we hold that generic aiding and abetting does not require "full knowledge" because the requirement arose in the context of an atypical statute and numerous jurisdictions have refused to apply it elsewhere.

Third, and relatedly, Alfred has failed to demonstrate that Washington's accomplice liability scheme does not reflect generic law. Washington's accomplice liability scheme is not unique. The Washington Supreme Court's decision in Davis, for example, is not an outlier: many states allow conviction of an accomplice to armed robbery without proof that the accomplice knew another person would brandish a firearm. See, e.g., State v. Pond, 315 Conn. 451, 108 A.3d 1083, 1099 (2015); Jones v. State, 648 So.2d 1210, 1211 (Fla. Dist. Ct. App. 1995); People v. Overten, 28 Cal.App.4th 1497, 34 Cal. Rptr. 2d 232, 234 (1994). And as we have explained, Washington is not the only state that allows conviction of an accomplice for knowing conduct. See, e.g., Ind. Code § 35-41-2-4. Accordingly, we cannot identify anything "special" about Washington's accomplice liability law, such that it would render Washington law overbroad. See Duenas-Alvarez, 549 U.S. at 191, 127 S.Ct. 815.

Struggling with our discussion of Washington law, the dissent cites State v. Dreewes, 192 Wash.2d 812, 432 P.3d 795 (2019), for the proposition that an accomplice in Washington can be convicted of assault with a deadly weapon despite lacking knowledge of the principal's intent to commit assault. Dissenting Op. at 1070-71. But Dreewes held no such thing. Instead, the Washington Supreme Court found that a jury could reasonably infer that the defendant, Jennifer Dreewes, knew that her compatriots would commit assault with a deadly weapon. Dreewes had "offered the coparticipants $300 to go to the identified residence, retrieve her items, give the pink-haired girl two black eyes, and abduct her and bring her to Dreewes's property." Dreewes, 432 P.3d at 801. Moreover, Dreewes "encouraged her coparticipants to arm themselves and informed them that four to five adults were at the residence." Id. at 802. The Washington Supreme Court held that this conduct supported a reasonable inference that despite not "know[ing] the names of all potential victims," Dreewes knew her actions would "promot[e] or facilitat[e] the crime eventually charged." Id. (internal quotation marks omitted and emphasis added). Such knowing conduct is sufficient to establish generic intent. See Rosemond, 572 U.S. at 77, 134 S.Ct. 1240.

The dissent's discussion of Dreewes leads to its core complaint—the dissent accuses us of "collapsing the purpose and knowledge mental states." Dissenting Op. at 1071. We have done no such thing. We recognize that knowledge and purpose are distinct. See Wash. Rev. Code § 9A.08.010(1)(a)-(b). We also do not dispute that some states have explicitly held that knowledge cannot support a conviction as an accomplice without additional proof of intent. See, e.g, Priddy v. Commonwealth, 629 S.W. 3d 14, 19 (Ky. App. 2021). Nevertheless, our task is to determine whether Alfred was convicted under a statute that "sweeps more broadly than the generic crime." Descamps, 570 U.S. at 261, 133 S.Ct. 2276. We have demonstrated that this is not the case; conduct that one knows would promote or facilitate the commission of the crime, like purposeful conduct, establishes intent to aid and abet under generic law. See Rosemond, 572 U.S. at 77, 134 S.Ct. 1240 ("[F]or purposes of aiding and abetting law, a person who actively participates in a criminal scheme knowing its extent and character intends that scheme's commission.").

Lastly, the concurrence argues that Washington departs from federal generic law because Washington allows conviction of an individual who "incidentally facilitate[d] a criminal venture rather than actively participate[d] in it." Concurring Op. at 1062 (quoting Rosemond, 572 U.S. at 77 n.8, 134 S.Ct. 1240). But the concurrence points to no case in which Washington permitted a conviction for "incident[al] facilitat[ion]." Id. On the contrary, numerous cases indicate that Washington requires accomplices to actively participate in the underlying crime. See, e.g., State v. Luna, 71 Wash.App. 755, 862 P.2d 620, 623 (1993) ("A defendant is not guilty as an accomplice unless he has associated with and participated in the venture as something he wished to happen and which he sought by his acts to make succeed." (emphasis added)); State v. Clemmons, No. 40847-3-II, 2012 WL 2366895, at *6 (Wash. Ct. App. June 22, 2012) (finding that a woman charged with rendering criminal assistance could not have known a man had committed murder because the government did not prove the woman knew that the man had "actively participated in the murders as an accomplice" (emphasis added)); State v. Gobena, No. 63174-8, 2010 WL 1541478 (Wash. Ct. App. Apr. 19, 2010) (upholding conviction as an accomplice "[b]ecause the State presented ample evidence establishing that [the defendant] was not only present at the crime scene but also actively assisting the principal" (emphasis added)). Indeed, in an opinion issued after Rosemond, the Washington Supreme Court could find "no Washington case upholding . . . liability . . . where the accused did not actively participate in the immediate physical impetus of harm." State v. Bauer, 180 Wash.2d 929, 329 P.3d 67, 73 (2014) (emphasis added). Thus, even if the concurrence is correct and "the knowledge and intent standards amount to the same thing" only in the "subset of aiding and abetting cases involving active participation in the actual criminal conduct," Concurring Op. at 1062, all Washington cases would fall into such a subset.

V. EFFECT ON PRECEDENT

In Valdivia-Flores, we concluded that Washington had "codifie[d] the distinction between intent and knowledge and ma[de] plain that knowledge is a less demanding mens rea requirement." 876 F.3d at 1207. We did not consider Rosemond, nor did we conduct any survey of generic accomplice liability law. After reviewing those sources, we have now explained why we disagree with Validivia-Flores's core premise that there is a material difference between intent and knowledge in this context. Rosemond's discussion of the intent requirement to prove federal aiding and abetting law emphasized the "advance knowledge" of the putative accomplice, 572 U.S. at 78, 134 S.Ct. 1240, and is consistent with Washington's required proof that the defendant "acted with knowledge that his or her conduct would promote or facilitate the crime [charged]," Cronin, 14 P.3d at 757. We thus overrule Valdivia-Flores.

To the extent that it remains good law, we overrule United States v. Franklin for the same reasons. 904 F.3d 793 (9th Cir. 2018), abrogated on other grounds by Shular v. United States, — U.S. —, 140 S. Ct. 779, 206 L.Ed.2d 81 (2020). Although Franklin reaffirmed Valdivia-Flores, albeit on broader grounds, it also failed to consider Rosemond. See Franklin, 904 F.3d at 799 ("if we also look outside federal law to define generic aiding and abetting liability for purposes of the ACCA, we reach the same result as under Valdivia-Flores's narrower, federal-law-centered, approach"). Franklin concluded that Washington was an outlier because it allowed conviction of an accomplice who merely knew "that a crime is being committed or is about to be committed." Franklin, 904 F.3d at 799 (internal quotation marks omitted) (emphasis added). As we have pointed out repeatedly, this is an incorrect characterization of Washington law. See Cronin, 14 P.3d at 758.

VI. CONCLUSION

We hold that second-degree robbery under Wash. Rev. Code § 9A.56.190 is a categorical match with generic theft. Accordingly, we conclude that Alfred has committed an aggravated felony and deny his petition for review.

PETITION DENIED.

COLLINS, Circuit Judge, with whom CALLAHAN and BADE, Circuit Judges, join as to Parts I and II, and with whom VANDYKE, Circuit Judge, joins as to Part III, concurring in the judgment in part and dissenting in part:

The question in this case is whether an alien's conviction for second-degree robbery in violation of Washington Revised Code §§ 9A.56.190 and 9A.56.210, and for which the sentence of imprisonment was more than one year, counts as an "aggravated felony" rendering the alien removable under § 237(a) of the Immigration and Nationality Act ("INA"). I agree with the votes of a majority of the court that the answer to that question is yes. And because Petitioner McKenzy Alii Alfred was convicted of second-degree robbery under those Washington statutes in 2018 and was sentenced to 15 months in prison on that charge, he was convicted of an "aggravated felony" and was properly found to be removable on that basis. But I disagree with, and therefore dissent from, most of the reasoning in the lead opinion. I concur only in the judgment.

I

Section 237(a) of the INA provides that "[a]ny alien . . . in and admitted to the United States shall, upon the order of the Attorney General, be removed if the alien is within one or more" of certain enumerated "classes of deportable aliens." See 8 U.S.C. § 1227(a). Among these classes of deportable aliens are those who have committed any of the various "[c]riminal offenses" described in § 237(a)(2). Id. § 1227(a)(2) (heading). In particular, § 237(a)(2)(A)(iii) provides that "[a]ny alien who is convicted of an aggravated felony at any time after admission" is "deportable." Id. § 1227(a)(2)(A)(iii). Section 101(a)(43) of the INA, in turn, contains a very lengthy definition of the term "aggravated felony" that includes 21 main alternative categories of criminal offenses. See id. § 1101(a)(43). Under the final sentences of § 101(a)(43), any offense falling within one of these 21 categories counts as an "aggravated felony" regardless of "whether in violation of Federal or State law" and "regardless of whether the conviction was entered before, on, or after" September 30, 1996. See 8 U.S.C. § 1101(a)(43); see also Pub L. No. 104-208, Div. C, § 321(b), 110 Stat. 3009, 3009-628 (Sept. 30, 1996).

The relevant category at issue here is set forth in § 101(a)(43)(G), which defines as an aggravated felony "a theft offense (including receipt of stolen property) . . . for which the term of imprisonment [is] at least one year." See 8 U.S.C. § 1101(a)(43)(G). Determining whether the "term of imprisonment" is "at least one year" is easy enough—one needs only to look at the relevant documents showing the sentence and judgment in the criminal case. See Alberto-Gonzalez v. INS, 215 F.3d 906, 910 (9th Cir. 2000) (holding that "the phrase 'for which the term of imprisonment [is at least] one year' " in § 101(a)(43)(G) "refer[s] to the actual sentence imposed by the trial judge"). Given that the sentence for Alfred's robbery conviction was 15 months, this requirement of § 101(a)(43)(G) is plainly satisfied here. The only remaining question is whether that robbery conviction counts as one for a "theft offense."

Because removability under the INA turns on whether Alfred has been "convicted," 8 U.S.C. § 1227(a)(2)(A)(iii) (emphasis added), of a "theft offense," id. § 1101(a)(43)(G) (emphasis added), the statutory language plainly "predicate[s] deportation 'on convictions, not conduct,' " Mellouli v. Lynch, 575 U.S. 798, 805, 135 S.Ct. 1980, 192 L.Ed.2d 60 (2015) (citation omitted). As a result, our task is not to "examin[e] whether an individual's actions meet a federal standard . . . , but only whether the individual 'has. . . been convicted of an offense' that does so." Pereida v. Wilkinson, — U.S. —, 141 S. Ct. 754, 762, 209 L.Ed.2d 47 (2021) (citation omitted). Accordingly, the INA calls for a "categorical approach," which requires the court to determine "whether 'the state statute defining the crime of conviction' categorically fits within the 'generic' federal definition of a corresponding aggravated felony." Moncrieffe v. Holder, 569 U.S. 184, 190, 133 S.Ct. 1678, 185 L.Ed.2d 727 (2013) (citation omitted).

The INA, however, does not define the crucial phrase "theft offense." Accordingly, we must "interpret that phrase using the normal tools of statutory interpretation." Esquivel-Quintana v. Sessions, 581 U.S. 385, 391, 137 S.Ct. 1562, 198 L.Ed.2d 22 (2017). Given that " '[e]lements' are the 'constituent parts' of a crime's legal definition," Mathis v. United States, 579 U.S. 500, 504, 136 S.Ct. 2243, 195 L.Ed.2d 604 (2016) (citation omitted); see also Torres v. Lynch, 578 U.S. 452, 457, 136 S.Ct. 1619, 194 L.Ed.2d 737 (2016) ("The substantive elements primarily define the behavior that the statute calls a 'violation' of federal law" (simplified)), the INA's use of the phrase "theft offense" must be understood as referring to the elements of the generic concept of "theft," i.e., "the offense [of theft] as commonly understood," Descamps v. United States, 570 U.S. 254, 257, 133 S.Ct. 2276, 186 L.Ed.2d 438 (2013). Our task, then, under the categorical approach, "focus[es] solely on whether the elements of the crime of conviction sufficiently match the elements of generic [theft], while ignoring the particular facts of the case." Mathis, 579 U.S. at 504, 136 S.Ct. 2243 (emphasis added); see also Descamps, 570 U.S. at 257, 133 S.Ct. 2276 (stating that the categorical approach requires courts to "compare the elements of the statute forming the basis of the defendant's conviction with the elements of the 'generic' crime").

Applying this comparison-of-elements approach to second-degree robbery under Washington law makes this an easy case. We undertook exactly that inquiry in United States v. Alvarado-Pineda, 774 F.3d 1198 (9th Cir. 2014), where we held that "a conviction for Washington second-degree robbery, where accompanied by a sentence of at least one year, qualifies as an 'aggravated felony' " under § 101(a)(43)(G). Id. at 1203. Specifically, we held that the federal generic understanding of "theft" has four elements, namely, "(1) the taking of (2) property (3) without consent (4) with the intent 'to deprive the owner of rights and benefits of ownership.' " Id. at 1202 (citation omitted). Reviewing the applicable Washington law, we held that the elements of second-degree robbery under Washington Revised Code §§ 9A.56.190 and 9A.56.210 are "(1) taking (2) personal property (3) from another person or from another's immediate presence (4) against his or her will (5) by force or threatened force (6) with the specific intent to steal." Id. at 1202-03 (citation omitted). Having thus set forth the relevant elements that define both offenses, we found a categorical match inasmuch as all of the elements of federal generic theft are necessarily met if the elements of the more serious offense of Washington second-degree robbery are satisfied:

A comparison of the elements of the statute to the elements of generic theft

makes clear that the "full range of conduct" criminalized by Washington second-degree robbery falls within the meaning of generic theft in that both require (1) the taking of (2) personal property (3) without consent and (4) with the specific intent to steal. One leading treatise states that "[r]obbery consists of all [the] elements of larceny . . . plus two additional requirements": that the property be taken from the victim's presence, and that the taking be accomplished by means of force or fear.

The lead opinion agrees with Alvarado-Pineda on this point, and it also notes that "[n]either Alfred nor the government has questioned the correctness of that decision." See Opin. at 1031. In my view, that should be the end of the inquiry. As defined by its generic elements, the federal generic concept of a "theft offense" under § 101(a)(43)(G) unquestionably embraces the crime of Washington second-degree robbery, as that offense is defined by the elements set forth in § 9A.56.190. Because Alfred's sentence exceeded one year, that "theft offense" was an "aggravated felony." See 8 U.S.C. § 1101(a)(43)(G). Alfred therefore was "convicted of an aggravated felony," and he was properly found to be removable. Id. § 1227(a)(2)(A)(iii). On that basis, the petition for review should be denied.

II

I hasten to add that the straightforward analysis I have set forth above is contrary to our prior decision in United States v. Valdivia-Flores, 876 F.3d 1201 (9th Cir. 2017). There, we held that, in "determin[ing] whether an offense qualifies as an 'aggravated felony' " under the INA, the categorical approach requires us to go beyond simply comparing the elements of the generic federal offense and the particular state offense. Id. at 1206. We reasoned that, because "aiding and abetting liability" is "implicit" in "every criminal charge," the categorical approach requires us also to determine whether the "definition of aiding and abetting liability" under the relevant state's law "is essentially the same as the federal definition so that they do, in fact, match categorically." Id. at 1207 (emphasis added). Consequently, under Valdivia-Flores, a mismatch in aiding and abetting theories under state and federal law could be enough to find that the categorical approach has not been satisfied, even though the relevant state offense's elements are a categorical match to the elements of the particular federal generic "aggravated felony." For multiple reasons, I respectfully disagree with Valdivia-Flores on this point, and I would overrule that aspect of the decision.

A

Nothing in the language of the INA supports Valdivia-Flores's view that, in determining whether a particular state offense qualifies as an aggravated felony, courts must undertake a comparative analysis of state and federal concepts of aiding and abetting liability. On the contrary, consideration of the statutory text leads to the opposite conclusion.

As I have explained, the text of the INA requires the court to determine whether the alien has been "convicted of an aggravated felony," 8 U.S.C. § 1227(a)(2)(A)(iii), with the relevant felony here being a "theft offense," id. § 1101(a)(43)(G). Because crimes are defined by their elements, see Mathis, 579 U.S. at 504, 136 S.Ct. 2243, the INA's references to a particular "felony" or "offense" must be understood as referring only to the elements of the relevant generic federal crime or the relevant state offense. See supra at 1049-50. And that is exactly what the Supreme Court has stated, time and again. See, e.g., Mathis, 579 U.S. at 504, 136 S.Ct. 2243 (explaining that the categorical approach "focus[es] solely on whether the elements of the crime of conviction sufficiently match the elements" of the relevant generic offense (emphasis added)); id. at 510, 136 S.Ct. 2243 ("We consider only the elements of the offense, without inquiring into the specific conduct of this particular offender." (simplified) (emphasis added)). Indeed, the Court in Mathis aptly noted that this "simple point" has become a "mantra" in its categorical-approach decisions, including those "in immigration cases." Id. at 510 & n.2, 136 S.Ct. 2243.

Mathis further underscored this "elements-only" approach by drawing a sharp distinction between the "elements" of a prior offense and " 'alternative means' of commission" of an offense. Id. at 514, 136 S.Ct. 2243 (citation omitted). The "elements" of a prior offense tell us "what a jury 'necessarily found' to convict a defendant (or what he necessarily admitted)," because each of those elements must be unanimously found by the jury beyond a reasonable doubt. Id. at 515, 136 S.Ct. 2243 (emphasis added) (citation omitted); see also Ramos v. Louisiana, — U.S. —, 140 S. Ct. 1390, 1397, 206 L.Ed.2d 583 (2020); In re Winship, 397 U.S. 358, 364, 90 S.Ct. 1068, 25 L.Ed.2d 368 (1970). By contrast, an alternative "means" of committing an offense is a "non-elemental fact" that "is by definition not necessary to support a conviction." Mathis, 579 U.S. at 515, 136 S.Ct. 2243 (simplified). Thus, for example, Mathis noted that Iowa's burglary law required the "burgled location" to be a qualifying "premises," but it did not require the jury to agree which of the various types of "premises" the particular location was (e.g., a "building, other structure, or vehicle"). Id. at 517-18, 136 S.Ct. 2243. In that example, the alternative ways of satisfying the "premises" element were therefore "means" and not "elements." Id. Having drawn this clear distinction between elements and means, Mathis squarely held that the categorical approach "disregards the means by which the defendant committed his crime, and looks only to that offense's elements." Id. at 517, 136 S.Ct. 2243.

Of course, where, as in Mathis, the listed means (i.e., building, structure, or vehicle) are alternative methods of proving a particular element (i.e., premises), that overarching element subsumes those alternative means, and in that sense those means will still be relevant in ascertaining the sweep of that element in undertaking the categorical approach's elements-only comparison. Thus, in Mathis, the inclusion of "vehicles" in the Iowa burglary statute's premises element meant that that element did not categorically match the corresponding element of the federal generic offense, which "requires unlawful entry into a building or other structure." Id. at 507, 136 S.Ct. 2243 (citation and internal quotation marks omitted). But the requisite categorical analysis under Mathis remains strictly an "elements-only inquiry." Id. at 510, 136 S.Ct. 2243.

Mathis's reasoning refutes Valdivia-Flores's view that the categorical approach requires us to undertake a comparison of state and federal aiding and abetting theories. It is well settled, as a matter of both federal law and Washington law, that aiding and abetting is merely a means of finding principal liability, that it does not itself constitute an element of the underlying offense, and that jurors need not unanimously agree as to whether a defendant is guilty as a principal or as an aider and abettor. See United States v. Garcia, 400 F.3d 816, 818-20 (9th Cir. 2005) (holding that, because "[a]iding and abetting is simply one means of committing a single crime," it "does not matter whether some jurors found that [the defendant] performed these acts himself, and others that he intended to help someone else who did"); State v. Hoffman, 116 Wash.2d 51,804 P.2d 577, 605 (1991) (holding that Washington law does not require "that jurors be unanimous as to the manner of an accomplice's and a principal's participation as long as all agree that they did participate in the crime"). When aiding and abetting is invoked, it does not displace the elements of the underlying offense, and the jury must still find that, as to either or both of the co-participants, each of the elements of that offense have been met. See, e.g., United States v. Mann, 811 F.2d 495, 497 (9th Cir. 1987) (noting that, even when the Government relies on an aiding and abetting theory, it "must prove that someone committed the underlying crime"); see also United States v. Sutcliffe, 505 F.3d 944, 959 (9th Cir. 2007) ("[A] defendant can only be convicted for aiding and abetting where some underlying crime has been committed."); State v. Carothers, 84 Wash.2d 256, 525 P.2d 731, 736 (1974) (noting that, when aiding and abetting is invoked, "[t]he elements of the crime remain the same"), overruled on other grounds by State v. Harris, 102 Wash.2d 148, 685 P.2d 584, 587 (1984); State v. Slater, 3 Wash.App. 615, 476 P.2d 719, 721 (1970) (holding that, when the prosecution relies on aiding and abetting, it is still "required to prove that a crime had actually been committed"); see generally WAYNE R. LAFAVE, SUBSTANTIVE CRIMINAL LAW § 13.3(c) (3d ed. 2022 update) (stating that "the guilt of the principal must be established at the trial of the accomplice as a part of the proof on the charge against the accomplice"). These considerations underscore that the requirements of aiding and abetting liability do not qualify as "elements" of the underlying offense for purposes of the categorical analysis. See Mathis, 579 U.S. at 517-18, 136 S.Ct. 2243 (stating that, if under the applicable substantive law, the "jury need not agree" on a particular point, that point is not an "element" for purposes of the categorical approach). It follows that, under Mathis, we must "disregard[ ]" aiding and abetting liability in determining whether there is a categorical match between a state offense and a federal generic offense. Id. at 517, 136 S.Ct. 2243.

Moreover, unlike in Mathis, the issue of aiding and abetting liability is not subsumed into one of the constituent elements of the underlying predicate offense. In Mathis, the particular range of means at issue (i.e., building, structure, or vehicle) involved alternative options for satisfying the premises element of the Iowa burglary statute. As a result, they were still relevant to the categorical approach's elements-only analysis, because they necessarily informed the understanding of the substantive breadth of that premises element. See supra at 1051-53. The same cannot be said of aiding and abetting liability. When aiding and abetting has been invoked as a theory of liability with respect to a particular defendant, it requires the prosecution to show that (1) the underlying offense was committed—i.e., each element of the offense is satisfied as to one or both of the co-participants; and (2) the particular defendant is liable for that offense under the relevant jurisdiction's aiding and abetting principles. See supra at 1052-53. Because the categorical approach focuses only on the elements of the charged offense (viz., component (1) listed above), it disregards aiding and abetting liability (viz., component (2) listed above). As Mathis explained, "Congress well knows" how to require consideration of "non-elemental facts," namely, by using "different language" from a mere reference to a type of offense. 579 U.S. at 511, 136 S.Ct. 2243. Here, there is no such distinct language in the INA that could be construed as referring to any relevant "non-elemental facts," such as whether the defendant's liability as a principal rests on aiding and abetting. Congress is presumed to know the law, see Guerrero-Lasprilla v. Barr, — U.S. —, 140 S. Ct. 1062, 1072, 206 L.Ed.2d 271 (2020), and it therefore was well aware that every jurisdiction uses some form of aiding and abetting liability as an alternative means of principal liability for any given offense. But Congress nonetheless used no language suggesting that it intended the INA's applicability to turn on whether a conviction was or was not based on aiding and abetting, much less that it turns on what particular version of aiding and abetting liability a jurisdiction employs.

The plurality nonetheless seems to think that, when aiding and abetting liability is invoked, the requirements of that doctrine (or at least the mental state requirement) effectively serve as "elements" for purposes of the categorical approach and therefore must be considered. According to the plurality, "because accomplice liability is one means of satisfying an element necessary to support a conviction, we must consider it." Opin. at 1035. The plurality's premise appears to be that, because the elements of the underlying offense might be proved only as to the third-party perpetrator and not as to the defendant, the requirements of the aiding and abetting doctrine that renders the defendant liable for the perpetrator's offense effectively serve as the "elements" of the defendant's offense. Although there may be certain contexts in which the plurality's analogy makes sense, it remains inapplicable to the categorical approach called for by the INA, which focuses only on whether the particular crime that is listed on the defendant's judgment of conviction qualifies as a "theft" offense. As I have explained, that inquiry calls only for a comparison of the elements of that theft offense to the elements of federal generic theft. See supra at 1049-52. The simple language Congress used—i.e., whether the defendant has been "convicted," 8 U.S.C. § 1227(a)(2)(A)(iii), of a "theft offense," id. § 1101(a)(43)(G)—ignores the subject of aiding and abetting and instructs us to ask whether the particular "offense" of conviction listed on the defendant's conviction documents—here, second-degree robbery under Washington Revised Code §§ 9A.56.190 and 9A.56.210—is a "theft" offense. Because offenses are defined by their elements, that would naturally be understood as an instruction to examine the elements of second-degree robbery and to compare them to the elements of generic theft. Congress did not ask us to go beyond examining the particular offense of conviction and to determine whether (due to the availability of aiding and abetting liability) the underlying conduct that led to the defendant's conviction may not itself have constituted "theft." See Descamps, 570 U.S. at 263, 133 S.Ct. 2276 ("[T]he categorical approach's central feature [is] a focus on the elements, rather than the facts, of a crime.").

In nonetheless insisting that an elements-based approach embraces aiding and abetting, the plurality reasons that the general availability of the alternative of aiding and abetting liability is equivalent to the availability of alternative means for proving a given element within the definition of a particular offense. Thus, in the plurality's view, (1) the ability to invoke aiding and abetting (with its allegedly lesser mens rea) is no different from (2) a hypothetical statute in which the prosecution may prove the underlying theft offense itself through either of two alternative mental states. See Opin. at 1035-36. Treating these two circumstances differently, the plurality concludes, "makes no sense." See Opin. at 1036. But this is just another way of saying that Congress should have drawn a line that gives controlling weight to the defendant's underlying conduct, rather than to what the particular offense of conviction happens to be. Congress chose the latter line, and so the focus is limited to a comparison of the elements of the particular offense of conviction. Options for proving one of the constituent elements within the definition of a particular offense must be considered in that elements-only analysis, but the general availability of aiding and abetting liability is not relevant.

Noting that the elements-only approach allows a court to determine "what facts are necessarily established by a conviction for the state offense," Moncrieffe, 569 U.S. at 205 n.11, 133 S.Ct. 1678 (emphasis added), the plurality worries that, absent an additional inquiry into state aiding and abetting law, a court would be unable to draw any conclusions as to what the individual defendant's underlying conduct necessarily was. See Opin. at 1036-37. According to the plurality, the defendant's conviction of a "theft offense" might just mean that someone else whom the defendant aided committed a theft offense and not that the defendant's conduct actually amounted to theft. See Opin. at 1036. Once again, the plurality erroneously assumes that the statute's focus is on defining and delimiting the defendant's personal underlying conduct. But that is not what the statute says. As I have repeatedly explained, it merely asks whether the defendant has been "convicted," 8 U.S.C. § 1227(a)(2)(A)(iii), of a "theft offense," id. § 1101(a)(43)(G). That simple language merely instructs the court to ascertain whether the particular offense that is listed on the alien's conviction documents counts as a "theft offense" under the familiar elements-only approach. It does not call for the court to "traverse the metaphysical line between" principals and accomplices "as a way of distinguishing degrees of culpability." See Opin. at 1034.

Accordingly, in determining whether a given state criminal statute qualifies as a "theft offense" within the meaning of the INA, all that matters is whether the elements of the state offense are a categorical match for federal generic "theft." Principles of aiding and abetting liability do not constitute elements of that underlying offense and are therefore irrelevant.

B

Valdivia-Flores nonetheless appeared to assume, without discussion, that a categorical-match inquiry into aiding and abetting theories was required by Gonzales v. Duenas-Alvarez, 549 U.S. 183, 189, 127 S.Ct. 815, 166 L.Ed.2d 683 (2007). See Valdivia-Flores, 876 F.3d at 1207. The plurality's opinion makes this claim explicitly (as does Judge VanDyke in his dissent). See Opin. at 1031-34; see also VanDyke Dissent at 1071-74. But Duenas-Alvarez did no such thing.

In Duenas-Alvarez, the Supreme Court addressed whether we had been correct in "holding that 'aiding and abetting' a theft is not itself a crime that falls within the generic definition of theft" for purposes of INA § 101(a)(43)(G). See 549 U.S. at 188-89, 127 S.Ct. 815. The Court concluded that we had erred and that the "generic definition of theft" under § 101(a)(43)(G) does include "one who aids or abets a theft" either "during [or] before the crime." Id. at 189-90, 127 S.Ct. 815. As the Court explained, because the laws of the federal Government and of all 50 States treat aiders and abettors as equivalent to principals, the INA's reference to a generic "theft offense" necessarily "covers such 'aiders and abettors' as well as principals." Id. at 190, 127 S.Ct. 815. In other words, the "criminal activities" of such "aiders and abettors of a generic theft" must be understood as "themselves fall[ing] within the scope of the term 'theft' in the federal statute." Id. Nothing in the analysis set forth earlier is inconsistent with Duenas-Alvarez's holding that the generic federal "theft offense" includes a theft offense that may have rested on an aiding and abetting theory. On the contrary, by "disregard[ing]" such "non-elemental" considerations as aiding and abetting liability, the categorical approach's "elements-only inquiry" ensures that the term "theft offense" will cover any aiders and abettors who have been convicted of a qualifying "theft offense" under the elements test. Mathis, 579 U.S. at 510, 515, 517, 136 S.Ct. 2243.

The plurality, however, concludes that a categorical-match comparison of aiding and abetting theories is required as a result of the Supreme Court's analysis of Duenas-Alvarez's alternative argument challenging the particulars of California's approach to aiding and abetting. See Opin. at 1031-32. That is wrong. In the cited discussion, the Court addressed Duenas-Alvarez's contention that, even conceding that the phrase "theft offense" must ordinarily be understood as including aiders and abettors of theft, "California's version of the doctrine" of aiding and abetting was so "unlike that of most other States"—so "special," in the Court's phrasing—that "it criminalizes conduct that most other States would not consider 'theft.' " Duenas-Alvarez, 549 U.S. at 191, 127 S.Ct. 815. Taking Duenas-Alvarez's argument at face value, the Court proceeded to reject it on the straightforward ground that its predicate assumption was wrong: Duenas-Alvarez, the Court held, had simply failed to "show that California's law is somehow special." Id. Nothing in the Court's holding or analysis on this point suggests that it was even undertaking—much less mandating—a categorical-match analysis between federal generic aiding and abetting principles and California aiding and abetting principles. The Court conspicuously did not undertake to identify, as the plurality does, the elements of "generic accomplice liability" and to then compare them to the elements of the relevant State's aiding and abetting liability. See Opin. at 1036-47. It simply rejected, as unsupported, Duenas-Alvarez's underlying premise that California was some sort of extreme outlier on aiding and abetting doctrine. See Duenas-Alvarez, 549 U.S. at 193, 127 S.Ct. 815 (noting that, even if California's aiding and abetting caselaw arguably contained some "expansive concept[s]," those concepts did not "extend significantly beyond the concept as set forth in the cases of other States"). Moreover, because the Court rejected that premise, the Court never addressed what would have followed if Duenas-Alvarez had been correct in his analysis of California law. Put simply, Duenas-Alvarez cannot bear the weight the plurality places on it.

C

There are additional considerations that further confirm that the categorical approach does not require us to undertake a comparative analysis of federal and state aiding and abetting principles.

First, the practical implications of Valdivia-Flores's approach are so nonsensical that we should not attribute them to Congress unless the statutory language requires that result (which it does not). In Valdivia-Flores, we found a categorical mismatch between Washington aiding and abetting liability and generic federal aiding and abetting liability, and we noted that one potential consequence of finding that mismatch was that "no Washington state conviction can serve as an aggravated felony at all." 876 F.3d at 1209 (emphasis added). We nonetheless rejected the Government's "pragmatic argument" that it "cannot have been Congress's intent" to thus exempt all Washington felons from this aspect of immigration law, because we concluded that, as an "inferior court," we were bound to accept that consequence of the categorical approach. Id. at 1208-09. However, for the reasons I have explained, nothing in Supreme Court precedent requires us to undertake a categorical-match analysis of aiding and abetting theories as an aspect of the categorical approach. And I find it altogether impossible to believe that, by simply instructing us to determine whether aliens have convictions for such offenses as "illicit trafficking in a controlled substance," "illicit trafficking in firearms," a "theft offense," a "burglary offense," "murder," "forgery," and "perjury," 8 U.S.C. § 1101(a)(43)(A), (B), (C), (G), (R), (S), Congress was supposedly thereby instructing us to assess the adequacy of each State's aiding and abetting theories and then, if we found them inadequate, to grant state-wide exemptions from those provisions of the INA to all criminal aliens in the relevant States. Surely if Congress had meant to establish such a general regime of wholesale state-by-state exemption from the criminal alien provisions of the INA, it would have used language more clearly indicative of such an extraordinary result. See AMG Cap. Mgmt., LLC v. FTC, — U.S. —, 141 S. Ct. 1341, 1349, 209 L.Ed.2d 361 (2021) ("Congress does not hide elephants in mouseholes." (simplified)).

Second, Valdivia-Flores's approach would produce a very peculiar anomaly within the INA itself. Most of INA § 101(a)(43)'s alternative "aggravated felony" definitions turn on a whole-offense comparison of elements of the sort described above with respect to the "theft offense" here. But there are some in which an offense can be designated as an "aggravated felony" based on the presence of a single element. For example, one enumerated category of "aggravated felony" is a "crime of violence," which is defined as "an offense that has as an element the use, attempted use, or threatened use of physical force against the person or property of another." 18 U.S.C. § 16(a) (emphasis added) (cross-referenced in INA § 101(a)(43)(F)). Because, in such a case, the inquiry turns only on whether the state offense contains that one particular element, no other feature of the state offense (including its applicable theory of aiding and abetting liability) is relevant. See Amaya v. Garland, 15 F.4th 976, 985-86 (9th Cir. 2021) (holding that Valdivia Flores's comparison of aiding and abetting theories does not apply when the element-of-physical-force clause of the definition of "crime of violence" is invoked (citing United States v. Door, 917 F.3d 1146 (9th Cir. 2019)). The result would be that we would not be required to undertake a categorical-match comparison of aiding and abetting theories in determining whether an alien is deportable for having committed a "crime of violence," but (under Valdivia-Flores) we would have to undertake such a comparison in the case of convictions involving "illicit trafficking in a controlled substance," "illicit trafficking in firearms," a "theft offense," a "burglary offense," "murder," "forgery," and "perjury." 8 U.S.C. § 1101(a)(43)(A), (B), (C), (G), (R), (S). It is extremely unlikely that Congress intended such a peculiar disparity in approach when it created its list of aggravated felonies in INA § 101(a)(43).

Third, there is no reason why Valdivia-Flores's (and the plurality's) expansion of the categorical approach to include alternative theories of principal liability would be limited to aiding and abetting. Next up, we will presumably have to consider whether a given State's version of Pinkerton liability matches federal generic Pinkerton liability and whether a state's version of causing-a-violation liability matches a federal generic version. Once again, it seems hard to believe that, merely by instructing us to determine whether Alfred's judgment of conviction was for a "theft offense," Congress was thereby requiring us to undertake such a comparative analysis of every alternative general theory for criminal liability in a given jurisdiction. The more sensible conclusion is that the INA ignores all such background principles of principal liability and only asks us to check whether, in light of its elements, the particular offense listed on a given judgment qualifies as a "theft offense."

* * *

For all of these reasons, I would overrule Valdivia-Flores to the extent that it requires a comparison of state and federal aiding and abetting theories. The categorical approach's "elements-only inquiry" requires us to disregard such non-elements and to instead compare only the elements of the state offense and the federal generic offense. Because it is undisputed that Alvarado-Pineda correctly found such a categorical match between the elements of Washington second-degree robbery and the elements of federal generic theft, Alfred was convicted of a "theft offense." Because his sentence for that offense exceeded one year, his conviction constitutes an "aggravated felony" and he was properly found to be deportable and removable. His petition for review must therefore be denied.

III

What I have said thus far is sufficient to dispose of the case, and I therefore am not required to address whether Washington aiding and abetting law is a categorical match for federal generic aiding and abetting law. While I therefore ultimately decline to address that specific issue, I nonetheless feel compelled to note that, on the way to its conclusion that Washington aiding and abetting law matches federal generic aiding and abetting law, the majority makes several statements about the scope of federal aiding and abetting law under 18 U.S.C. § 2 that are contrary to well-settled authority.

A

Before addressing the majority's serious misstatements about federal aiding and abetting under § 2, I cannot help making a few preliminary observations about the peculiar way in which the majority approaches its task of defining generic aiding and abetting liability for purposes of applying the categorical approach.

The majority's articulation of federal generic aiding and abetting liability largely ignores the conduct requirements and instead focuses on the mental state requirement. As to that issue, the majority notes that the States follow at least four different approaches—some require "purpose"; others require only "knowledge"; others use the same mens rea as for the underlying offense; and others employ the so-called "natural and probable consequences doctrine." See Opin. at 1038-39. Noting that "a majority of jurisdictions use statutory language that requires a showing of purpose or something similar," the majority "hold[s] that generic accomplice liability requires a showing that the putative accomplice intentionally aided or abetted another in the commission of the crime." See Opin. at 1039, 1041-42 (emphasis added). But because the majority deems the terms "intent," "purpose," and "knowledge" to be "largely synonymous" in the "context" of aiding and abetting, the majority "also conclude[s] that advance knowledge of the crime is sufficient to support a conviction for generic accomplice liability." See Opin. at 1039, 1041-42. And, that means, according to the majority, that the generic form of aiding and abetting is "broad" and "encompass[es] jurisdictions that require purposeful conduct, as well as those that require accomplices to act with knowledge that their conduct will promote or facilitate commission of the crime." See Opin. at 1042. Moreover, based on the Supreme Court's decision in Duenas-Alvarez, the majority concludes that the "generic definition of accomplice liability" also "encompasses jurisdictions that use a natural and probable consequences doctrine." See Opin. at 1042. And, of course, the derivative mens rea approach will generally require at least knowledge (which, as the majority notes, roughly corresponds to "the concept of general intent" that is ordinarily the minimum mens rea required for most offenses). See Opin. at 1040. As a result, all of the States' disparate approaches to mens rea are embraced within the majority's expansive concept of "generic accomplice liability."

Thus, in contrast to the typical categorical approach to defining a given generic offense—in which a specific set of minimum elements would be articulated—the majority ultimately does not come up with a single formula for generic aiding and abetting. Instead, the majority concocts a multiple-alternatives approach that includes every variety of the mens rea requirement that is currently in use. Given that, by the end of section IV(B) of its opinion, the majority has already defined generic aiding and abetting so broadly and so eclectically that it includes all of the approaches followed by all 50 States, I am not quite sure why it takes so many additional pages to find a categorical match with Washington law.

Ironically, the end result of the majority's expansive approach to defining generic aiding and abetting liability is arguably the same as mine. That is, if (as the majority posits) "generic aiding and abetting" is so broadly defined that every alternative approach qualifies as a match, then the ultimate outcome is the same as it would be if (as I have argued) the entire step of comparing aiding and abetting theories is simply omitted altogether. While I am gratified with that bottom line result, the majority's analysis nonetheless is predicated on a seriously flawed intermediate premise. Specifically, as I explain in the next section, the majority is quite wrong in stating that, in the context of aiding and abetting liability, purpose and knowledge are "largely synonymous."

B

In finding a categorical match between Washington and federal generic aiding and abetting law, the majority makes two key errors that directly bear on the scope of federal aiding and abetting liability under 18 U.S.C. § 2. First, the majority contends that the construction of § 2 aiding and abetting in Rosemond v. United States, 572 U.S. 65, 134 S.Ct. 1240, 188 L.Ed.2d 248 (2014), supports the majority's erasure of the distinction between intent and knowledge in the context of aiding and abetting generally. Second, the majority also dramatically narrows the applicability of Rosemond's holding that, to qualify as aiding and abetting under § 2, the defendant's "inten[t] to facilitate th[e] offense's commission" "must go to the specific and entire crime charged." Id. at 76, 134 S.Ct. 1240. The majority seriously errs on both counts. And because we are sitting en banc, the majority's errors have now overturned settled law in ways that may have a significant impact across the entire federal criminal justice system in our circuit.

1

To set the majority's errors in context, I begin with some important background concerning the scope of aiding and abetting liability under § 2.

One of the central problems of aiding and abetting law is how to define the line between genuinely culpable assistance in the commission of a particular offense and more peripheral or attenuated conduct that might be said merely to indirectly facilitate an offense. This concern is reflected in the many hypotheticals mentioned on this subject in the internal debates of the American Law Institute ("ALI") that preceded its adoption of the Model Penal Code. Would aiding and abetting liability extend, for example, to a "utility [that] provides telephone or telegraph services, knowing it is used for bookmaking" or to a "vendor [who] sells with knowledge that the subject of the sale will be used in the commission of a crime"? As reflected in the ALI's discussions, there are at least two ways to address this crucial line-drawing problem. One approach, which was ultimately adopted in the Model Penal Code, is to cabin aiding and abetting liability by requiring proof that the defendant acted with a highly culpable mental state. Thus, that Code allows one who "aids" in the commission of an offense to be liable as a principal only if he or she does so "with the purpose of promoting or facilitating the commission of the offense." See MODEL PENAL CODE § 2.06(3)(a)(ii) (ALI 1985) (emphasis added). A second approach, which was reflected in an earlier draft of the Code, combined a less demanding mental state of knowledge with more "rigorously drafted" "conduct requirements." See id. cmt. 6(c) (emphasis added). Under this approach, it would suffice to show that the aider and abettor "knowingly facilitated" the commission of a crime, but the requisite facilitation had to be "substantial" or to have involved "provid[ing] the means or opportunity for the commission of the crime." Id. (emphasis added).

The federal aiding and abetting statute, 18 U.S.C. § 2, has long been construed as following the first approach—that is, it combines a demanding mental state requirement with a more broadly defined facilitation requirement. As the Supreme Court explained in Rosemond, the aider and abettor must be shown to have "intend[ed] to facilitate [the] offense's commission" in the sense that it is " 'something that he wishes to bring about.' " 572 U.S. at 76-77, 134 S.Ct. 1240 (quoting Nye & Nissen v. United States, 336 U.S. 613, 619, 69 S.Ct. 766, 93 L.Ed. 919 (1949) (in turn quoting United States v. Peoni, 100 F.2d 401, 402 (2d Cir. 1938) (L. Hand, J.))). Moreover, it is not sufficient that the defendant intended "to advance some different or lesser offense"; "[i]nstead, the intent must go to the specific and entire crime charged." Id. at 76, 134 S.Ct. 1240 (emphasis added). But this demanding mental state requirement is paired with a fairly expansive "affirmative-act requirement." Id. at 74, 134 S.Ct. 1240. To establish the requisite facilitation of the offense, the Government only needs to show that the defendant's actions "facilitated one component" of the offense and the "importance of the aid rendered" is not "relevant." Id. at 74-75, 134 S.Ct. 1240. Indeed, the Court explained, "a person's involvement in the crime could be not merely partial but minimal too." Id. at 73, 134 S.Ct. 1240. Given the "minimal" nature of the affirmative-act requirement under federal aiding and abetting law, the demanding scienter requirement plays an "important" role in ensuring that the Government is not able to "sweep within the drag-net of conspiracy [and abetting] all those who have been associated in any degree whatever with the main offenders." United States v. Falcone, 109 F.2d 579, 581 (2d Cir. 1940) (L. Hand, J.).

2

The majority's decision today, however, wrongly weakens § 2's demanding mental state requirement in two key respects.

a

First, the majority erroneously contends that, under Rosemond's description of § 2 aiding and abetting liability, intent and knowledge are "largely synonymous," and that, as a result, there is "no significant daylight" between Washington's knowledge standard for aiding and abetting liability and § 2's intent standard as construed in Rosemond. See Opin. at 1040, 1043-44; see also Opin. at 1040-41, 1043-44. The majority claims that its across-the-board equivalence between knowledge and intent in the context of aiding and abetting follows from Rosemond's holding that "a person who actively participates in a criminal scheme knowing its extent and character intends that scheme's commission." 572 U.S. at 77, 134 S.Ct. 1240 (emphasis added). The majority's reasoning fails, because it ignores the load-bearing significance of the above-italicized language from Rosemond.

As I have explained, see supra at 1061-62, Rosemond unambiguously endorses "Judge Learned Hand's" "canonical formulation of th[e] needed state of mind" for criminal aiding and abetting liability, under which it must be shown that the defendant had the intent to "bring about" the crime and "by his action to make it succeed." Id. at 77 (citation omitted). However, in surveying the relevant caselaw, the Court observed that in cases involving a defendant who "actively participates in a criminal adventure," the defendant's "full knowledge of the circumstances constituting the charged offense" will suffice to "satisf[y]" the "intent requirement." Id. at 77 (emphasis added). That makes sense, because if the defendant is personally and actively involved in the actual carrying out of the crime with full and "advance" knowledge of what is contemplated, that defendant unquestionably intends for that crime to be accomplished. Id. at 78. Rosemond thus observed that, in this subset of aiding and abetting cases involving active participation in the actual criminal conduct, the knowledge and intent standards amount to the same thing: "for purposes of aiding and abetting law, a person who actively participates in a criminal scheme knowing its extent and character intends that scheme's commission." Id. at 77 (emphasis added).

But the Court immediately added a footnote cautioning that this conclusion rested dispositively on the premise that the defendant's affirmative act of assistance extended beyond the minimal conduct that is ordinarily sufficient for aiding and abetting purposes and instead involved active participation in the crime:

We did not deal in these cases, nor do we here, with defendants who incidentally facilitate a criminal venture rather than actively participate in it. A hypothetical case is the owner of a gun store who sells a firearm to a criminal, knowing but not caring how the gun will be used. We express no view about what sort of facts, if any, would suffice to show that such a third party has the intent necessary to be convicted of aiding and abetting.

In addition to finding no support in Rosemond, the majority's wholesale collapsing of knowledge and intent for aiding and abetting purposes is plainly incorrect. The majority's premise that knowledge and intent are equivalent for aiding and abetting purposes would have come as a surprise to the drafters of the Model Penal Code, who debated over the critical differences between those standards in aiding and abetting law. As explained earlier, those drafters explicitly recognized that, if knowledge were to be adopted as the standard rather than intent, "the conduct requirements" would have to be "more rigorously drafted" in order to avoid giving a novel and unwarranted sweep to aiding and abetting principles. See MODEL PENAL CODE § 2.06, cmt. 6(c) (emphasis added). That recognition coheres with Rosemond's observation that, in cases of active participation in the crime (which is a situation in which the "conduct requirement[ ]" has been more "rigorously" articulated), there is no practical difference between knowledge and intent. But the same cannot be said for cases relying on the full breadth of the conduct requirement under federal law, under which "a person's involvement in the crime could be not merely partial but minimal too." Rosemond, 572 U.S. at 73, 134 S.Ct. 1240.

In Falcone, Judge Learned Hand similarly recognized the inappropriateness of combining a less demanding mental state of knowledge with a broad understanding of what conduct counts as aiding and abetting. Surveying the circuit split that had developed over the knowledge-intent issue in cases involving criminal prosecutions against persons who supplied materials to illegal distilleries, Judge Hand explained as follows why the Second Circuit had come down on the side of requiring intent:

[I]n prosecutions for conspiracy or abetting, [the defendant's] attitude towards the forbidden undertaking must be more positive. It is not enough that he does not forego a normally lawful activity, of the fruits of which he knows that others will make an unlawful use; he must in some sense promote their venture himself, make it his own, have a stake in its outcome. The distinction is especially important today when so many prosecutors seek to sweep within the drag-net of conspiracy all those who have been associated in any degree whatever with the main offenders. That there are opportunities of great oppression in such a doctrine is very plain, and it is only by circumscribing the scope of such all comprehensive indictments that they can be avoided. We may agree that morally the defendants at bar should have refused to sell to illicit distillers; but, both morally and legally, to do so was toto coelo different from joining with them in running the stills.

The majority asserts that its across-the-board equating of aiding and abetting liability ultimately makes no practical difference here, because Washington law supposedly pairs its lower knowledge requirement for aiding and abetting with an active-participation conduct requirement. See Opin. at 1046-47. That proposition seems doubtful as a matter of Washington law, but more importantly, it does nothing to assuage the concerns raised by the majority's broader equating of knowledge and intent in the context of aiding and abetting. As noted earlier, the majority's discussion of generic aiding and abetting entirely ignores the conduct requirement of aiding and abetting, but it nonetheless flatly says—based in part on Rosemond—that knowledge and intent are "largely synonymous." See Opin. at 1040. And in footnote 10 of its opinion, the majority makes clear that the "blurry" line between "incidental facilitation and active participation" makes no difference to its conclusion. See Opin. at 1047 n.10. For the reasons I have explained, Rosemond does not support the majority's across-the-board equating of intent and knowledge in the context of aiding and abetting.

b

Second, the majority wrongly limits a critical aspect of Rosemond's description of aiding and abetting liability under § 2.

Although Washington law requires that the defendant act "with knowledge that he or she was promoting or facilitating the crime for which that individual was eventually charged," State v. Cronin, 142 Wash.2d 568, 14 P.3d 752, 758 (2000) (emphasis in original), it does not require that the defendant's knowledge extend to the "entire crime charged," Rosemond, 572 U.S. at 76, 134 S.Ct. 1240 (emphasis added). Thus, under Washington law, a defendant who knowingly abets a robbery may be convicted of armed robbery, even if he or she was unaware that the robber would be armed. See State v. Davis, 101 Wash.2d 654, 682 P.2d 883, 884-85 (1984). By contrast, in Rosemond, the Court held that, under § 2, the defendant could not be convicted of aiding and abetting the use of a gun during a drug trafficking crime in violation of 18 U.S.C. § 924(c) unless it was shown that he "intended the commission" of "an armed drug sale." 572 U.S. at 78, 134 S.Ct. 1240 (emphasis added).

The conflict between Davis and Rosemond on this point is so patent that the majority was only able to eliminate it by dramatically limiting the scope of Rosemond's holding. According to the majority, Rosemond's requirement of advance knowledge of the entire crime (such as the fact that it was to be an armed offense) applies only to a narrow class of "combination" crimes that are closely analogous to § 924(c). See Opin. at 1044-45. Section 924(c) qualifies as such a "combination" crime, in the majority's view, because it "is a 'freestanding offense' that requires proof of two distinct acts," whereas, "[i]n contrast, first-degree armed robbery is an enhanced version of a base offense—simple robbery." See Opin. at 1045. Because Rosemond's "entire crime" rule does not apply to the mine run of cases, the majority reasons, there is no disparity between Washington and federal aiding and abetting law on this point. All of this is wrong.

As an initial matter, the majority ignores the fact that, even before Rosemond, we have long recognized and applied the same rule that, "[t]o be convicted as an aider and abettor, the defendant must have knowingly and intentionally aided and abetted the principals in each essential element of the crime." United States v. Dinkane, 17 F.3d 1192, 1196 (9th Cir. 1994) (emphasis added); see also United States v. Short, 493 F.2d 1170, 1172 (9th Cir. 1974) (explaining that this rule follows from general common law principles). In Dinkane, we applied this same entire-crime-charged rule (later endorsed in Rosemond) to hold that, "[i]n order to convict a defendant for armed bank robbery under an aiding and abetting theory," the government must "show beyond a reasonable doubt," inter alia, "that the defendant knew that the principal had and intended to use a dangerous weapon during the robbery." 17 F.3d at 1195 (emphasis added). That, of course, is the exact opposite of Washington aiding and abetting law under Davis. The majority's effort to limit Rosemond's "entire crime charged" rule to "combination crimes" is thus directly contrary to Dinkane.

Moreover, the majority's posited distinction between a "combination crime" and an "enhanced version of a base offense" is illusory and ultimately incoherent. Section 924(c) makes it a criminal offense, inter alia, to use or carry a firearm "during and in relation to any crime of violence or drug trafficking crime." 18 U.S.C. § 924(c)(1)(A). As a result, all of the elements of the predicate crime of violence or drug trafficking crime must be proved, together with the additional element that a firearm was used "during and in relation to that predicate offense." Id.; see also United States v. Mendoza, 25 F.4th 730, 741-42 (9th Cir. 2022). In that respect, a § 924(c) offense that is based on using a gun during a bank robbery is analytically indistinguishable from an armed bank robbery charge: both offenses require proof of all of the elements of the predicate/base offense of bank robbery plus an additional respective element relating to the use of a firearm. See 18 U.S.C. §§ 924(c)(1)(A), 2113(a), (d). In Blockburger terms, a bank robbery offense is every bit as much of a lesser-included offense of a bank-robbery-based § 924(c) charge as it is of an armed bank robbery charge. See Blockburger v. United States, 284 U.S. 299, 304, 52 S.Ct. 180, 76 L.Ed. 306 (1932) (holding that the test for whether two offenses are distinct for double jeopardy purposes is "whether each provision requires proof of a fact which the other does not"). The only difference is that Congress has explicitly allowed a § 924(c) count to be separately charged and punished in addition to the predicate offense. See 18 U.S.C. § 924(c)(1)(D)(ii); see also United States v. Davis, 306 F.3d 398, 417-18 (6th Cir. 2002).

Given that, in terms of the constituent elements of the offenses, a "combination crime" and an "enhanced version of a base offense" have exactly the same relationship to the underlying predicate/base offense, there is no principled basis for saying that Rosemond's "entire crime" rule applies to one but not the other. 572 U.S. at 76, 134 S.Ct. 1240. Moreover, nothing in Rosemond suggests that the Court was articulating a special mental state rule that applies, as the majority would have it, only in the context of a peculiar subset of "combination crimes." On the contrary, in holding that the intent required for aiding and abetting "must go to the specific and entire crime charged," the Rosemond Court described that holding as a "general rule" that was based on longstanding common law principles. See id. at 76 & n.7, 134 S.Ct. 1240. Indeed, Rosemond cited with approval two cases that (like Dinkane) held that, as the Court put it, "the unarmed driver of a getaway car had the requisite intent to aid and abet armed bank robbery if he 'knew' that his confederates would use weapons in carrying out the crime." Id. at 77, 134 S.Ct. 1240 (emphasis added) (citing United States v. Akiti, 701 F.3d 883, 887 (8th Cir. 2012); United States v. Easter, 66 F.3d 1018, 1024 (9^th Cir. 1995)).

* * *

At best, the majority's en banc misreading of Rosemond casts a serious cloud over the previously settled scope of federal aiding and abetting under § 2. At worst, it arguably overturns decades of settled law on that subject. I respectfully dissent from that aspect of the majority's opinion.

IV

For the foregoing reasons, I concur in the majority's judgment that the petition should be denied, but I dissent from essentially all of its reasoning.

CALLAHAN, Circuit Judge, with whom Bade, Circuit Judge, joins, concurring in the judgment:

I vote to deny the petition. Because we need not consider aiding and abetting liability, I concur in Sections I and II of Judge Collins' concurrence in part and dissent in part. However, because by the vote of a majority of the en banc panel aiding and abetting liability remains before us, I also concur in subsections B and C of Section IV and Section V of Judge Bybee's opinion and agree that Washington's aiding and abetting law is not overbroad.

McKEOWN, Circuit Judge, with whom MURGUIA, Chief Judge, and S.R. THOMAS and VANDYKE, Circuit Judges, join, dissenting:

In yet another case, we are called on to conduct what sounds like a simple analysis—"whether '[Washington's] statute defining the crime of conviction' categorically fits within the 'generic' federal definition of a corresponding aggravated felony." Moncrieffe v. Holder, 569 U.S. 184, 190, 133 S.Ct. 1678, 185 L.Ed.2d 727 (2013) (quoting Gonzales v. Duenas-Alvarez, 549 U.S. 183, 186, 127 S.Ct. 815, 166 L.Ed.2d 683 (2007)). But that proposition is easier stated than applied. Consequently, we have said that "perhaps no other area of the law has demanded more of our resources." United States v. Aguila-Montes de Oca, 655 F.3d 915, 917 (9th Cir. 2011) (en banc) (opinion of Bybee, J.), abrogated on other grounds by Descamps v. United States, 570 U.S. 254, 133 S.Ct. 2276, 186 L.Ed.2d 438 (2013). Because "[t]he categorical approach requires us to perform absurd legal gymnastics," Lopez-Aguilar v. Barr, 948 F.3d 1143, 1149 (9th Cir. 2020) (Graber, J., concurring), it is no surprise that scholars have asked whether this is a "categorical approach or categorical chaos." Other circuits have observed that the categorical approach "push[es] us into a catechism of inquiry that renders these approaches ludicrous." United States v. Williams, 898 F.3d 323, 337 (3d Cir. 2018) (Roth, J., concurring). Because courts are required to parse numerous state statutes in applying immigration and criminal law, often leading to inconsistent conclusions, it is no wonder "the categorical approach has developed a reputation for crushing common sense." United States v. Escalante, 933 F.3d 395, 406 (5th Cir. 2019).

Notwithstanding such obvious frailties, the categorical approach forces us into the straitjacket of slicing and dicing the statutes, often producing "arbitrary and inequitable results." Mathis v. United States, 579 U.S. 500, 521, 136 S.Ct. 2243, 195 L.Ed.2d 604 (2016) (Kennedy, J., concurring). But since we must do so, it is only fair to adhere to state law as state law is written, not as we wish it to be, and to land on a generic definition that comports with the Model Penal Code, the majority of state statutes, and leading treatises. Because the majority takes a path that diverges from these principles, I respectfully dissent.

I agree with the majority that, under the categorical approach, we must compare the state statute to the federal generic definition of the offense. I also agree that because Washington's statutory scheme incorporates accomplice liability into all crimes, we must consider accomplice liability in our categorical analysis of Washington second-degree robbery. But I part ways with the majority in two significant respects. First, we differ on the generic definition of accomplice liability. A close read of the relevant sources reveals that generic accomplice liability requires a mental state of purpose, which is different than knowledge. Second, we disagree on whether Washington's second-degree robbery statute is a categorical match with the generic theft offense. Washington accomplice liability requires a mental state of knowledge, which is lower than purpose. Because the Washington law is therefore a categorical mismatch with generic theft, I would grant Alfred's petition for review.

The majority recites the relevant considerations for discerning the generic definition of accomplice liability before concluding that it is "difficult to discern a clear generic standard." Yet, the majority continues with apparent confidence, because the purpose mental state is interchangeable with the knowledge mental state, "advance knowledge of the crime is sufficient to support a conviction for generic accomplice liability." The legal landscape, however, is far less chaotic than the majority suggests, and the majority's cited authorities do not support this conclusion. Indeed, generic accomplice liability requires a mental state of purpose, not knowledge.

To begin, the Model Penal Code expressly distinguishes purpose and knowledge. The Model Penal Code defines four culpable mental states: (1) purposely, (2) knowingly, (3) recklessly, and (4) negligently. Model Penal Code § 2.02(2) (Am. L. Inst. 2021). Specifically:

(a) Purposely. A person acts purposely with respect to a material element of an offense when:

(i) if the element involves the nature of his conduct or a result thereof, it is his conscious object to engage in conduct of that nature or to cause such a result; and

(ii) if the element involves the attendant circumstances, he is aware of the existence of such circumstances or he believes or hopes that they exist.

(b) Knowingly. A person acts knowingly with respect to a material element of an offense when:

(i) if the element involves the nature of his conduct or the attendant circumstances, he is aware that his conduct is of that nature or that such circumstances exist; and

(ii) if the element involves a result of his conduct, he is aware that it is practically certain that his conduct will cause such a result.

Having differentiated between purpose and knowledge, the Model Penal Code states that a person is liable as an accomplice if:

(a) with the purpose of promoting or facilitating the commission of the offense, he

(i) solicits such other person to commit it, or

(ii) aids or agrees or attempts to aid such other person in planning or committing it, or

(iii) having a legal duty to prevent the commission of the offense, fails to make proper effort so to do; or

(b) his conduct is expressly declared by law to establish his complicity.

In general, "[t]he generic definition of an offense roughly corresponds to the definitions of the offense in a majority of the States' criminal codes." United States v. Garcia-Jimenez, 807 F.3d 1079, 1084 (9th Cir. 2015) (internal quotation marks and citation omitted). As the majority acknowledges, twenty-six state criminal codes comport with the Model Penal Code and require that an accomplice act with a mental state of purpose or intent. See 2 Wayne R. LaFave, Substantive Criminal Law § 13.2(b), at 466-67 nn.70, 72 (3d ed. 2018); see also, e.g., Ga. Code Ann. § 16-2-20(b); Ariz. Rev. Stat. § 13-301; Mont. Code Ann. § 45-2-302(3). The federal accomplice liability statute similarly requires a mental state of purpose. See 18 U.S.C. § 2 (providing that "[w]hoever commits an offense against the United States or aids, abets, counsels, commands, induces or procures its commission" and "[w]hoever willfully causes an act to be done which if directly performed by him or another would be an offense against the United States, is punishable as a principal"); United States v. Hernandez-Orellana, 539 F.3d 994, 1006 (9th Cir. 2008) (explaining that federal aiding and abetting requires "that the accused had the specific intent to facilitate the commission of a crime by another"). The majority admits as much by noting that "18 U.S.C. § 2(a) adequately reflects federal generic accomplice liability law."

Here, the generic accomplice liability standard is clear: intent is required. This conclusion is amplified by criminal-law treatises, which instruct that accomplice liability generally requires a higher mental state than knowledge. LaFave explains that, in general, "accomplice liability exists when the accomplice intentionally encourages or assists, in the sense that his purpose is to encourage or assist another in the commission of a crime as to which the accomplice has the requisite mental state." LaFave, supra, § 13.2(b), at 467. Further, although some states like Washington allow for accomplice liability based on "encouragement or assistance with knowledge that it will promote or facilitate a crime," accomplice "liability has seldom been imposed on this basis." Id. at 468-69; see also Jens David Ohlin, Wharton's Criminal Law § 10:9 (16th ed. 2022) (explaining that federal courts, the Model Penal Code, and some states follow the "purpose standard" rather than "the more relaxed knowledge standard").

Instead of relying on the consensus of twenty-six states, the majority leaps to conclude that purpose and knowledge are equivalent. In doing so, the majority focuses on five states—Wisconsin, Missouri, Idaho, Michigan, and Illinois—that ostensibly blur the line between purpose and knowledge. Even if five states' policies were a sufficient rebuttal to majority consensus, these state statutes do not support the majority's conclusion with the force the majority asserts. For example, Wisconsin, Missouri, and Idaho do not treat purpose and knowledge as interchangeable. Wisconsin law provides that intent is a required element of accomplice liability and can be proven with a combination of knowledge and either an overt act of assistance or an indication of willingness to assist. See State v. Hibbard, 404 Wis.2d 668, 982 N.W.2d 105, 110 (App. 2022); Wis. IJ-Criminal, No. 400. Missouri and Idaho, on the other hand, stipulate that either intent or knowledge can satisfy the mental state requirement. See, e.g., State v. Barker, 442 S.W.3d 165, 169 (Mo. Ct. App. 2014); State v. Gonzalez, 134 Idaho 907, 12 P.3d 382, 384 (App. 2000). And Michigan and Illinois have identified knowledge and intent as two separate factors that may support an inference of accomplice liability. See, e.g., People v. Robinson, 475 Mich. 1, 715 N.W.2d 44, 48 (2006); People v. Batchelor, 171 Ill.2d 367, 216 Ill.Dec. 519, 665 N.E.2d 777, 780-81 (1996). Allowing for multiple options of proving an element does not make those options equivalent.

The majority also cites Rosemond v. United States, 572 U.S. 65, 134 S.Ct. 1240, 188 L.Ed.2d 248 (2014), as evidence of the blurred line between purpose and knowledge. The Supreme Court's analysis in Rosemond, however, reveals that the purpose mental state in the federal aiding and abetting statute, 18 U.S.C. § 2, is higher than Washington's knowledge requirement. The Court in Rosemond, considering the conduct necessary to demonstrate that a person intended to facilitate the commission of a crime, held that "an aiding and abetting conviction requires not just an act facilitating one or another element, but also a state of mind extending to the entire crime." 572 U.S. at 75-76, 134 S.Ct. 1240 (emphasis added). The Court clarified that its holding was "grounded in the distinctive intent standard for aiding and abetting." Id. at 81 n.10, 134 S.Ct. 1240. Accordingly, as the majority acknowledges, a federal conviction for aiding and abetting requires the defendant to have had "full knowledge of the circumstances constituting the charged offense." See id. at 77, 134 S.Ct. 1240.

An accomplice in Washington need only participate in a crime with general knowledge of the crime being committed, not specific knowledge of "every element" of the crime. State v. Roberts, 142 Wash.2d 471, 14 P.3d 713, 736 (2000). Under Rosemond, for an accomplice to have intended "a [criminal] scheme's commission, he" must have participated in the scheme while understanding its "extent and character." 572 U.S. at 77, 134 S.Ct. 1240. Rosemond did not confirm the interchangeability of intent and knowledge. Instead, Rosemond clarified that advance knowledge is necessary, but not sufficient, to establish the purpose mental state.

The majority doubly errs by forcing its view of what Washington law should be and by collapsing the purpose and knowing mental states, an error highlighted by Washington law's express differentiation between those mental states. Washington's criminal code provides that "[a] person acts with intent or intentionally when he or she acts with the objective or purpose to accomplish a result which constitutes a crime." Wash. Rev. Code. § 9A.08.010(1)(a). A person acts with knowledge, however, when:

(i) He or she is aware of a fact, facts, or circumstances or result described by a statute defining an offense; or

(ii) He or she has information which would lead a reasonable person in the same situation to believe that facts exist which facts are described by a statute defining an offense.

Following Roberts, Washington has repeatedly articulated that knowledge of the principal's crime is sufficient to support a conviction for accomplice liability. See, e.g., State v. Berube, 150 Wash.2d 498, 79 P.3d 1144, 1151 (2003) (citing cases and noting that, "[f]or Berube's conviction as an accomplice to stand, the evidence must support a finding that she . . . kn[ew] that her acts would either promote or facilitate the crime"); State v. Carter, 154 Wash.2d 71, 109 P.3d 823, 829 (2005) (explaining that an erroneous jury instruction on accomplice liability was harmless when the jury's conviction was clearly based on the defendant's knowledge of the principal's charged crimes). The Washington Supreme Court has also held that a defendant has "adequate knowledge" for accomplice liability when he "acts with knowledge that [his] conduct will promote the specific crime charged." State v. Farnsworth, 185 Wash.2d 768, 374 P.3d 1152, 1159 (2016) (citing State v. Cronin, 142 Wash.2d 568, 14 P.3d 752, 759 (2000)).

This principle was reinforced when the Washington Supreme Court recently upheld a conviction for second-degree assault on an accomplice liability theory where the defendant argued that she lacked knowledge that her codefendants would assault the victim "with a deadly weapon." State v. Dreewes, 192 Wash.2d 812, 432 P.3d 795, 799 (2019). The court noted that the defendant knew generally of the plan to rob a home and kidnap someone there, even though she did not know that her co-defendant would assault the victim—who was not the person the co-defendant sought to kidnap—with a rifle. Id. at 802. The defendant's "general knowledge" of the substantive crime sufficed for her conviction, and specific knowledge of all the elements of the crime was not necessary. Id.

Further, in practice, an accomplice in Washington need not "have the intent" to commit a specific crime, "just knowledge that his actions were facilitating the crime." State v. A.L.Y., No. 56645-8-I, 2006 WL 2723983, at *3 (Wash. Ct. App. Sept. 25, 2006) (per curiam). In State v. A.L.Y., the defendant argued that he could not be guilty of second-degree robbery, even as an accomplice, because he lacked intent to steal, which is "an essential non-statutory element of robbery in Washington." Id. The Washington Court of Appeals clarified that an accomplice need not "share the same mental state as the principal," and that A.L.Y.'s general "knowledge that his actions were facilitating" the robbery was sufficient to support his conviction. Id. (quoting Berube, 79 P.3d at 1151).

Despite the clear instruction that Washington requires only knowledge for accomplice liability, the majority concludes that Washington's second-degree robbery statute is a categorical match with a generic theft definition that requires an accomplice to act with the purpose to aid and abet in the commission of a crime. The majority, unwilling to accept what Washington law is, imputes its own view of what Washington law should be.

Beyond wrongly concluding that Washington's statute is a categorical match with generic theft, the majority errs by collapsing the purpose and knowledge mental states. By the Model Penal Code's own definitions, purpose and knowing are different standards. Compare Model Penal Code § 2.02(2)(a), with § 2.02(2)(b); see Bailey, 444 U.S. at 405, 100 S.Ct. 624. The majority of states follow this scheme, but a state's decision to diverge from these definitions should not be a basis to elide those concepts when determining the generic offense, nor an excuse to rewrite clear state law. See Sarausad v. Porter, 503 F.3d 822, 831 (9th Cir. 2007) (Callahan, J., dissenting from denial of rehearing en banc) ("As a federal court, we should not expand our duties to rewrite state laws.").

My disagreement with the majority illustrates the confusion and frustration engendered by the categorical approach. As explained above, under the overwhelming weight of authority, there is an unambiguous definition of accomplice liability that requires a mental state of purpose or intent, not simply knowledge. But the majority sees it otherwise, declaring that the generic definition of accomplice liability is murky and then transforming that murkiness into an absolute definition that disfavors Alfred.

I respectfully dissent and would grant Alfred's petition.

VANDYKE, Circuit Judge, dissenting:

While I would like to join Judge Bybee's characteristically well-written decision, I agree with Judge Collins and my dissenting colleagues that the analysis therein incorrectly elides the distinction between the mental states of knowledge (or general intent) and purpose (or specific intent) to find a categorical match in this case. The distance between those mental states is too great a gulf to span. And while the approach taken by Judge Collins in his decision has a lot to commend it as an original matter, I also can't join it entirely because I don't believe his approach is ultimately reconcilable with what the Supreme Court actually did in Duenas-Alvarez. So I must dissent from our court's result in this case.

I write separately from the other dissenters only to address more specifically why I am prevented from fully joining Judge Collins's decision. Judge Collins characterizes as superfluous the Supreme Court's analysis of a disputed issue it actually decided in Duenas-Alvarez: whether California's doctrine of aiding and abetting was "special" and thus unlike the generic counterpart. 549 U.S. at 190-91, 127 S.Ct. 815. This was the main issue litigated by Duenas-Alvarez before the Supreme Court, but by characterizing it as a mere "alternative argument," Judge Collins minimizes what was really the heart of that case.

According to Judge Collins, "[n]othing in the Court's holding or analysis on this point suggests that it was even undertaking—much less mandating—a categorical-match analysis between federal generic aiding and abetting principles and California aiding and abetting principles." Respectfully, I don't think that is a permissible reading of Duenas-Alvarez. If Judge Collins was correct in suggesting the Court thought categorical analysis of aiding and abetting was unnecessary, the Court could have resolved section III(A) of Duenas-Alvarez in one easy sentence: "Because aiding and abetting is not part of the categorical analysis inquiry, Duenas-Alvarez's argument that California's aiding and abetting law is different from its generic counterpart is irrelevant." But that's not what the Court did. The Court said that " [t]o succeed, Duenas-Alvarez must show something special about California's version of the [aiding and abetting] doctrine—for example, that California in applying it criminalizes conduct that most other States would not consider 'theft.' " Duenas-Alvarez, 549 U.S. at 191, 127 S.Ct. 815 (first emphasis added). "Duenas-Alvarez attempt[ed] to make just such a showing," but the Supreme Court rejected it, not because it was irrelevant, but because Duenas-Alvarez failed in his attempt. Id. at 191-93, 127 S.Ct. 815. And the Court spent multiple pages explaining why. Id. at 190-93, 127 S.Ct. 815. If Judge Collins was correct that aiding and abetting is irrelevant to the categorical match analysis, then this whole discussion by the Court was an inexplicable extended frolic and detour.

Judge Collins nonetheless argues that the Court did not necessarily require a categorical match analysis of aiding and abetting because it merely compared "the particulars of California's approach to aiding and abetting" to generic aiding and abetting. What Judge Collins artfully calls "the Supreme Court's analysis of Duenas-Alverez's alternative argument challenging the particulars of California's approach to aiding and abetting" was, in fact, a categorical match analysis of aiding and abetting. But while the Court in Duenas-Alvarez clearly did a categorical match analysis of aiding and abetting, Judge Collins puts decisive weight on the fact that the Court never explicitly stated that the analysis it conducted was necessary. As he sees it, by doing a categorical analysis but not saying it needed to do what it did, "the Court did not reach th[e] broader issue" of whether what it did was necessary.

Respectfully, I find that an odd way to read Duenas-Alvarez—the Supreme Court devoted the central part of its opinion spanning multiple pages to doing a categorical match analysis of aiding and abetting. And that's presumably why Judge Collins's interpretation of Duenas-Alvarez is so original. See, e.g., Bourtzakis v. U.S. Att'y Gen., 940 F.3d 616, 620-22 (11th Cir. 2019). Under that approach to reading Supreme Court precedents, if the Court wants lower courts to follow its precedents, I guess it must tell, not show.

That is not an interpretive approach I think any lower court could apply consistently. Consider, for example, the applicability of the First Amendment in some particular context. Assume that the Supreme Court in cases in that context analyzed whether there had been a First Amendment violation by applying some well-established First Amendment test (say, strict scrutiny), and concluded in every case that, on the unique facts of each case, there was no violation. But the Court simply did the analysis; it never actually said it was required. Would we then feel free to conclude the First Amendment was categorically inapplicable in that context, because (quoting Judge Collins) the Court had never expressly "sua sponte raise[d] the threshold question of whether any such [First Amendment analysis] was required"? No. I think we would safely assume that if the First Amendment was categorically inapplicable in that context, the Court would not have gone to all the trouble to perform a strict scrutiny analysis. Or at least we should assume so until the Court tells us differently. To conclude otherwise is to impose a "magic words" requirement on the Supreme Court that has no basis in common sense or practice.

Judge Collins's characterization of Duenas-Alvarez's aiding-and-abetting-mismatch argument as a mere "alternative argument" that didn't merit much attention from the Supreme Court is inconsistent with that case's procedural history. By the time the Supreme Court reviewed Duenas-Alvarez, that argument was the central controversy before to the Court. The Court originally granted certiorari in the case to consider and overrule our court's prior decision in Penuliar v. Ashcroft, 395 F.3d 1037, 1044-46 (9th Cir. 2005), which had misapplied the categorical approach to hold that because (i) California's law of theft criminalized aiding and abetting, and (ii) California's aiding and abetting, unlike the generic theft offense, didn't involve taking or controlling property, that therefore California's law swept more broadly than, and thus did not match, the federal theft offense. See Duenas-Alvarez, 549 U.S. at 188-89, 127 S.Ct. 815. The Court granted review in Duenas-Alvarez to correct the part of our court's categorical match analysis that determined aiding and abetting a theft was not a crime that fell within the federal generic definition of theft. Id. But after the Supreme Court granted certiorari, Duenas-Alvarez did not even contest that point. See id. at 190, 127 S.Ct. 815 ("Duenas-Alvarez does not defend the Ninth Circuit's position."). Given that procedural posture, the Court's uncontroversial reversal of Penuliar required little work: "The question before us is whether one who aids or abets a theft falls, like a principal, within the scope of th[e] generic definition. We conclude that he does." Id. at 189, 127 S.Ct. 815. Indeed, the Court barely spent three paragraphs on the question. Id. at 189-90, 127 S.Ct. 815.

Instead, the meat of the case before the Supreme Court was precisely the issue that Judge Collins downplays as a mere "alternative argument." Seeing the writing on the wall from the Court's grant of certiorari, Duenas-Alvarez conceded that the generic law "treats aiders and abettors during and before [a theft] the same way it treats principals." Id. at 190, 127 S.Ct. 815. But he argued that he should still win because, acknowledging that categorical analysis applied to aiding and abetting, California's doctrine was so dissimilar from the generic counterpart as to be "special." Id. at 191, 127 S.Ct. 815. The Court took nine paragraphs and two detailed appendices to conclude that it was not "special."

Again, if Judge Collins was correct that the Supreme Court really thought aiding and abetting liability was simply irrelevant to the categorical match analysis, why did the Court go to all the trouble to address Duenas-Alvarez's argument on the merits at all? It would have been easy enough (indeed, much simpler than what it did) to just say: "It's irrelevant." Instead, the Court concluded that, "in our view, to find that a state statute creates a crime outside the generic definition of a listed crime in a federal statute requires more than the application of legal imagination to a state statute's language." Id. at 193, 127 S.Ct. 815. Rather, "[i]t requires a realistic probability, not a theoretical possibility, that the State would apply its statute to conduct that falls outside the generic definition of a crime." Id. These statements about how to apply the categorical match analysis are sandwiched in the middle of the Court's comparison between California's approach to aiding and abetting and the generic approach.

Judge Collins has made a powerful argument for what the law maybe should be, and maybe the Supreme Court will adopt that argument in the future. But it's too difficult for me to reconcile that with what the Court actually did in Duenas-Alvarez. I thus reluctantly dissent.