Leval, Circuit Judge

Kathleen M. Sullivan, Quinn Emanuel Urquhart & Sullivan, LLP, New York, NY (Robert L. Raskopf, Todd Anten, Quinn Emanuel Urquhart & Sullivan, LLP, New York, NY; Michael A. Cheah, Vimeo, LLC, New York, NY & Rachel Herrick Kassabian, Quinn Emanuel Urquhart & Sullivan, LLP, Redwood Shores, CA on the brief), for Defendants-Appellants-Cross-Appellees.

Carter G. Phillips, Sidley Austin, LLP, Washington, DC (Russell J. Frackman, Marc E. Mayer, Mitchell Silberberg & Knupp LLP, Los Angeles, CA; Constantine L. Trela, Jr., Sidley Austin LLP, Chicago, IL; Christine Lepera, Jeffrey M. Movit, New York, NY, Kwaku A. Akowuah, Washington, DC on the brief), for Plaintiffs-Appellees-Cross-Appellants.

Matt Schruers, Computer & Communications Industry Association, Washington, DC, for amicus curiae Computer & Communications Industry Association.

Corynne McSherry, Vera Ranieri, Electronic Frontier Foundation, San Francisco, CA; Sherwin Siy, Public Knowledge, Washington, DC; Art Neill, Teri Karobonik, New Media Rights, San Diego, CA; Elizabeth Rosenblatt, Rebecca Tushnet, Organization for Transformative Works, New York, NY, Emma Llansó, Center for Democracy and Technology, Washington, DC, for amicus curiae Electronic Frontier Foundation, Organization for Transformative Works, The Center for Democracy and Technology, Public Knowledge, and New Media Rights.

Michael Elkin, Thomas P. Lane, Winston & Strawn LLP, New York, NY; Robert Turner, Yahoo!, Inc., Sunnyvale, CA; A. John P. Mancini, Mayer Brown LLP, New York, NY, for amici curiae Google Inc., Facebook, Inc., Yahoo!, Inc., and Microsoft, Inc.

Brian M. Willen, Wilson Sonsini Goodrich & Rosati, P.C., New York, NY; Sara E. Rowe, Wilson Sonsini Goodrich & Rosati, P.C., Palo Alto, CA, for amici curiae Pinterest, Inc., Tumblr, Inc., and Twitter, Inc.

Thomas G. Hentoff, Jessica S. Richard, Williams & Connolly LLP, Washington, DC, for amici curiae Recording Industry Association of America, Inc., American Association of Independent Music Inc., National Music Publishers' Association, Inc., The Harry Fox Agency, Inc., Authors Guild, Inc., Association of American Publishers, American Society of Composers, Authors and Publishers, Broadcast Music, Inc., American Federation of Musicians of the United States and Canada, and Screen Actors Guild-American Federation of Television and Radio Artists.

Andrew H. Bart, Jenner & Block LLP, New York, NY; Luke C. Platzer, J. Douglas Wilson, Jenner & Block LLP, Washington, DC, for amicus curiae The Copyright Alliance.

Before: LEVAL, HALL, LYNCH, Circuit Judges:

Opinion

Leval, Circuit Judge:

This is an interlocutory appeal on certified questions from rulings of the United States District Court for the Southern District of New York (Abrams, J .). interpreting the Digital Millennium Copyright Act of 1998 (“DMCA”). The DMCA establishes a safe harbor in § 512(c), which gives qualifying Internet service providers protection from liability for copyright infringement when their users upload infringing material on the service provider's site and the service provider is unaware of the infringement. 17 U.S.C. § 512(c). Defendant Vimeo, LLC is an Internet service provider, which operates a website on which members can post videos of their own creation, which are then accessible to the public at large. Plaintiffs are record companies and music publishing companies, which own copyrights in sound recordings of musical performances. Their complaint alleges that Vimeo is liable to Plaintiffs for copyright infringement by reason of 199 videos posted on the Vimeo website, which contained allegedly infringing musical recordings for which Plaintiffs owned the rights.

The district court ruled on motions for partial summary judgment addressed to whether Vimeo was entitled to the DMCA's safe harbor protections. As for videos that allegedly infringed pre-1972 sound recordings, the court ruled in Plaintiffs' favor on the theory that § 512(c)'s safe harbor absolves a service provider only from copyright liability based on the federal copyright statute, which does not apply to pre-1972 sound recordings, which are protected only by state copyright laws. With respect to post-1972 sound recordings (which all agree are protected by the DMCA's safe harbor when its conditions are met), the district court granted summary judgment to Vimeo as to 153 videos, mostly on the basis that Plaintiffs lacked evidence that Vimeo's employees had viewed them. The court rejected Plaintiffs' arguments that knowledge should be imputed to Vimeo by reason of its alleged general policy of willful blindness to infringement of sound recordings. And as for the remaining challenged videos that incorporated post-1972 sound recordings, the court denied summary judgment to either side, concluding that there was a question of material fact whether Vimeo possessed so-called “red flag” knowledge of circumstances that made infringement apparent, which would make Vimeo ineligible for the protection of the safe harbor under the terms of § 512(c). This interlocutory appeal focuses on three issues: (i) whether the safe harbor of § 512(c) applies to pre-1972 sound recordings; (ii) whether evidence of some viewing by Vimeo employees of videos that played all or virtually all of “recognizable” copyrighted songs was sufficient to satisfy the standard of red flag knowledge, which would make Vimeo ineligible for the DMCA safe harbor; and (iii) whether Plaintiffs have shown that Vimeo had a general policy of willful blindness to infringement of sound recordings, which would justify imputing to Vimeo knowledge of the specific infringements.

We affirm the district court's rulings in part and vacate in part. (i) On the first question—whether the safe harbor protects service providers from infringement liability under state copyright laws—we conclude it does and accordingly vacate the district court's grant of partial summary judgment to Plaintiffs on this question. (ii) As to whether some viewing by a service provider's employee of a video that plays all or virtually all of a recognizable copyrighted song is sufficient to establish red flag knowledge, disqualifying the service provider from the benefits of the safe harbor, we rule that, under the standard set forth in Viacom International, Inc. v. YouTube, Inc. , 676 F.3d 19, 26 (2012), it does not. We therefore remand for reconsideration of the various denials of summary judgment in Vimeo's favor. (iii) On whether Plaintiffs showed a general policy of willful blindness that disqualifies Vimeo from claiming protection of the safe harbor, we agree with the district court's ruling in Vimeo's favor.

BACKGROUND

I. The DMCA

“The DMCA was enacted in 1998 to implement the World Intellectual Property Organization Copyright Treaty and to update domestic copyright for the digital age.” Viacom , 676 F.3d at 26 (internal citations and quotation marks omitted). According to its legislative history, Title II, the Online Copyright Infringement Liability Limitation Act was designed to “clarif[y] the liability faced by service providers who transmit potentially infringing material over their networks,” S. Rep. No. 105-190, at 2 (1998), and in the process to “ensure[ ] that the efficiency of the Internet will continue to improve and that the variety and quality of services on the Internet will expand.” Id. The Senate Report expressed the view that “without clarification of their liability, service providers may hesitate to make the necessary investment in the expansion of the speed and capacity of the Internet.” Id . at 8. To that end, the DMCA established four safe harbors, codified at 17 U.S.C. § 512, which protect qualifying Internet service providers from liability for certain claims of copyright infringement. Viacom , 676 F.3d at 27. This case focuses on the safe harbor provided by § 512(c), which is supplemented by protections provided in § 512(m).

These portions of the statute undertake, through complex provisions, to establish a compromise, which, on the one hand, augments the protections available to copyright owners, and, on the other, insulates service providers from liability for infringements of which they are unaware, contained in material posted to their sites by users, so as to make it commercially feasible for them to provide valuable Internet services to the public. The Act augments the rights of copyright owners by establishing a notice-and-takedown regime. The notice-and-takedown regime requires a service provider, to preserve its eligibility for the safe harbor, to “expeditiously ... remove ... material that is claimed to be infringing,” or disable access to it, whenever the service provider (1) receives a notice of infringing material on the service provider's site or (2) otherwise becomes aware of the infringement or of circumstances making the infringement apparent. § 512(c)(1)(C), (A)(iii). The provisions favoring Internet service providers, first, immunize those that qualify for the statute's benefits from liability for copyright infringements posted by users on the providers' websites if the service providers are unaware of the infringements, and, second, expressly relieve them of any obligation to monitor the postings of users to detect infringements as a condition of qualifying for the safe harbor. Service providers, however, forfeit entitlement to the safe harbor if they fail to expeditiously remove the infringing material upon receipt of notification of the infringement or upon otherwise becoming aware of it.

The terms summarized above are set forth in the following statutory provisions:

(c)(1) In general.—A service provider shall not be liable for monetary relief, or [with certain exceptions] for injunctive

or other equitable relief, for infringement of copyright by reason of the storage at the direction of a user of material that resides on a system or network controlled or operated by or for the service provider, if the service provider—

(A)

(i) does not have actual knowledge that the material or an activity using the material on the system or network is infringing;

(ii) in the absence of such actual knowledge, is not aware of facts or circumstances from which infringing activity is apparent; or

(iii) upon obtaining such knowledge or awareness, acts expeditiously to remove, or disable access to, the material;

(B) does not receive a financial benefit directly attributable to the infringing activity, in a case in which the service provider has the right and ability to control such activity; and

(C) upon notification of claimed infringement as described in paragraph (3), responds expeditiously to remove, or disable access to, the material that is claimed to be infringing or to be the subject of infringing activity.

§ 512(c)(1)(A)-(C).

(m) Protection of privacy.—Nothing in this section shall be construed to condition the applicability of [the safe harbor of § 512(c) ] on—

(1) a service provider monitoring its service or affirmatively seeking facts indicating infringing activity, [with exceptions not relevant to this inquiry].

§ 512(m)(1).

II. Vimeo's Website

Vimeo has had great success as a site for the storage and exhibition of videos. Its Website hosts a wide array of home videos, documentaries, animation, and independent films. Founded in 2005, as of 2012 it hosted more than 31 million videos and had 12.3 million registered users in 49 countries. Approximately 43,000 new videos are uploaded to Vimeo each day. Users post videos onto the website without the intervention or active involvement of Vimeo staff, and Vimeo staff do not watch or prescreen videos before they are made available on the website. When a video is uploaded, it is automatically converted to Vimeo's format and stored on Vimeo's servers. Users can view the videos stored on Vimeo servers through a “streaming” process by visiting the website, and in many instances can download them.

All Vimeo users must accept its Terms of Service. These require, inter alia, that: users upload (1) only videos that they have created or participated in creating, and (2) only videos for which they possess all necessary rights and that do not infringe on any third party rights. Vimeo's “Community Guidelines” also provide content restrictions and information about its copyright policy. Every time a user uploads a video, the Website displays three rules: (1) “I will upload videos I created myself,” (2) “I will not upload videos intended for commercial use,” and (3) “I understand that certain types of content are not permitted on Vimeo.” Nonetheless, users have the technical ability to upload videos that do not comply with the rules.

Vimeo employs a “Community Team” of 16 employees to curate content. These employees identify some videos with a “like” sign, occasionally prepare commentary on a video, offer technical assistance to users, participate in forum discussions, and at times inspect videos suspected of violating Vimeo's policies. So far as we are aware, the record does not indicate that the videos as to which the district court denied summary judgment were inspected by the Community Team for the purpose of detecting infringement.

In order to upload a video to the Website, a user must register for a Vimeo account. Vimeo offers two forms of membership accounts—basic (free) and paid subscription services. It derives revenue from user subscription fees and advertising on its website—the vast majority of its revenue coming from subscription sales. Unless the posting user has limited access, any Internet user can view, download, and copy videos posted on the website for free.

A registered user has the ability to note her “likes” or comment on videos, subscribe to “groups” of users with common interests, and subscribe to or create “channels” of videos based on themes.

Vimeo uses multiple computer programs (“Mod Tools”) that assist its Community Team in locating and removing videos that may contain content that violates the Terms of Service. When videos and/or users are identified by one of these tools, Vimeo staff review them individually. Vimeo also enables users to “flag” videos that they believe violate the Terms of Service. Community Moderators evaluate the flagged content and decide whether or not to remove it. The flagging interface also explains how to submit a DMCA claim.

Between October 2008 and November 2010, Vimeo deleted at least 4,000 videos in response to takedown notices by copyright owners. On the three identified occasions in which Plaintiffs had sent Vimeo takedown notices, the district court found that Vimeo had responded “expeditious[ly].” Plaintiffs did not send takedown notices regarding the videos involved in this suit.

While it appears that Vimeo followed a practice of screening the visual content of posted videos for infringement of films, it did not screen the audio portions for infringement of sound recordings. Plaintiffs contend that this fact, together with statements made by Vimeo employees (found in emails), show indifference and willful blindness to infringement of recorded music, and that Vimeo has furthermore actively encouraged users to post infringing videos. Plaintiffs' evidence of such statements by Vimeo employees included the following:

• Dalas Verdugo, a “Community Director” at Vimeo, responded to a user's question that he “see[s] all the time at vime[o] videos, (for example Lip-dub) music being used that is copyrig[ht]ed, is there any problem with this?” by telling the user “[w]e allow it, however, if the copyright holder sent us a legal takedown notice, we would have to comply.”

• Blake Whitman, a member of Vimeo's Community Team, responded to a question regarding Vimeo's “policy with copyrighted music used as audio for original video content” by telling the user, “[d]on't ask, don't tell ;).”

• On another occasion, Whitman responded to a user who asked about using a Radiohead song in a posted video by writing, “We can't officially tell you that using copyright music is okay. But ....”

• Andrea Allen, a member of Vimeo's Community Team, received a message from a user providing a link to a video and stating, “I have noticed several people using copyrighted music on Vimeo. What do you do about this?” Allen forwarded the e-mail internally with the comment “[i]gnoring, but sharing.”

• In a response to an email asking whether a user would have copyright “issues” with adding the copyrighted song “Don't Worry, Be Happy” by Bobby McFerrin as the “soundtrack” to a home video, Allen responded: “The Official answer I must give you is: While we cannot opine specifically on the situation you are referring to, adding a third party's copyrighted content to a video generally (but not always) constitutes copyright infringement under applicable laws ... Off the record answer ... Go ahead and post it ....”

• In an e-mail sent to Whitman and Verdugo (and also to all@vimeo.com), Andrew Pile, the Vice President of Product and Development at Vimeo, wrote: “Who wants to start the felons group, where we just film shitty covers of these [Plaintiff EMI] songs and write ‘FUCK EMI’ at the end?”

III. Proceedings Below

Plaintiffs filed complaints on December 10, 2009, charging Vimeo with direct, contributory, and vicarious copyright infringement. The complaints identified 199 videos that included recordings of music in which Plaintiffs claimed rights. From March 3, 2011, until April 4, 2012, the case was stayed pending our court's decision in Viacom.

In May, 2012, Plaintiffs sought leave to amend their complaints to add more than 1,000 videos to the suit. The Court denied the request without prejudice, on the ground that the additional videos would “require reopening of discovery and delay the timely adjudication of the proposed summary judgment motions.” Id.

On September 7, 2012, Vimeo moved for summary judgment on the basis of § 512(c)'s safe harbor. On November 16, 2012, Plaintiffs cross-moved for partial summary judgment that Vimeo was ineligible for the safe harbor.

In a September 18, 2013 order, the district court granted partial summary judgment to Plaintiffs as to videos that allegedly infringed pre-1972 sound recordings. The court granted summary judgment to Vimeo under the safe harbor as to 136 videos on the basis that there was no evidence that Vimeo employees had observed them. As to videos for which there was evidence of some observation by Vimeo employees, the court denied both sides' motions, ruling that there were triable issues of fact regarding whether Vimeo had acquired actual or red flag knowledge of infringement that would disqualify it from safe harbor protection. Id. The district court rejected Plaintiffs' argument that Vimeo should be held liable under a willful blindness theory.

On Vimeo's motion for reconsideration, the district court granted Vimeo summary judgment on an additional 17 videos, on 15 because of insufficient evidence of observation by Vimeo staff, and on two because they contained only short portions of the allegedly infringed recordings, which the court found insufficient to support a finding of red flag knowledge.

The court then certified two questions for interlocutory appeal: “(a) Whether the DMCA's safe-harbor provisions are applicable to sound recordings fixed prior to February 15, 1972”; and (b) “Whether, under the holding of Viacom, a service provider's viewing of a user-generated video containing all or virtually all of a recognizable, copyrighted song may establish ‘facts or circumstances' giving rise to ‘red flag’ knowledge of infringement.” This court granted Vimeo's petition for interlocutory review of those two questions, and granted Plaintiffs' request for interlocutory review of a third issue: whether Plaintiffs' evidence showed willful blindness that could justify imposition of liability on Vimeo, notwithstanding the safe harbor provisions.

DISCUSSION

I. Pre-1972 Recordings

The first question we consider is whether the district court erred in granting partial summary judgment to Plaintiffs, rejecting the availability of the DMCA's safe harbor for infringement of sound recordings fixed prior to February 15, 1972. (For convenience, we use the terms “pre-1972” and “post-1972” to refer to sound recordings fixed before, or after, February 15, 1972.) The district court concluded that, with respect to sound recordings, the safe harbor established by § 512(c) protects only against liability under the federal copyright law, and that the DMCA consequently gives service providers no protection for pre-1972 recordings, which are protected only by state laws of copyright.

Confusion on this issue results from Congress's convoluted treatment of sound recordings. Although sound recordings have existed since the 19th century, for reasons not easily understood Congress first included them within the scope of federal copyright protection on February 15, 1972, and the grant of federal copyright protection to sound recordings on that date applied only to sound recordings to be made thereafter. United States Copyright Office, Federal Copyright Protection for Pre -1972 Sound Recordings 5 (2011), available at http://copyright.gov/docs/sound/pre-72-report.pdf [“Pre-1972 Sound Recordings Report”]. Pre-1972 recordings have never been covered by the federal copyright. Accordingly, copyright protection of pre-1972 sound recordings has depended on the copyright laws of the states. Id.

In 1976, Congress enacted an overall revision of the law of copyright. Section 301 of the new statute, in subsection (a), asserted federal preemption (ousting all state laws) with respect to works covered by the federal copyright. The preemption did not include pre-1972 sound recordings as these were not covered by the federal copyright. Subsection (c) of § 301 provided with respect to pre-1972 sound recordings that “any rights or remedies under the common law or statutes of any State shall not be annulled or limited by this title until” February 15, 2047. After that date, federal law would preempt state law, so that state laws of copyright previously protecting pre-1972 sound recordings would cease to have effect, and all pre-1972 sound recordings would pass into the public domain. See Pub. L. No. 94-553, § 301, 90 Stat. 2541, 2572 (1976) (codified at 17 U.S.C. § 301 ). Subsequently, when Congress extended the duration of the federal copyright term, it passed parallel amendments to § 301(c), which similarly extended the period during which pre-1972 sound recordings would continue to be protected by state copyright laws. Section 301(c) in its present form postpones the date at which pre-1972 sound recordings will pass into the public domain until February 15, 2067—95 years after February 15, 1972. Pub. L. No. 105-298, 112 Stat. 2827 (1998). Plaintiffs argued in the district court, with success, and argue again on this appeal, that the interrelationship of § 301(c) with the safe harbor provision of § 512(c) requires that the latter be interpreted to have no application to pre-1972 sound recordings. Section 512(c), the safe harbor, provides that service providers meeting the qualifications of the statute “shall not be liable ... for infringement of copyright.” Plaintiffs argue that, if this safe harbor provision is interpreted to protect service providers from infringement liability under state copyright laws, it conflicts irreconcilably with § 301(c)'s provision that, until 2067, “rights or remedies under the common law or statutes of any State shall not be annulled or limited by this title.” According to Plaintiffs' argument, the proper way to reconcile § 301(c) with § 512(c), so as to avoid the conflict, is to construe § 512(c)'s guarantee that service providers “shall not be liable ... for infringement of copyright” as meaning that they shall not be liable for infringement of the federal copyright, but as having no application to any liability service providers may incur under state laws.

On this question, the district court accepted without discussion the position taken by the United States Copyright Office in a report prepared in 2011 that the safe harbor does not protect against liability for infringement of pre-1972 sound recordings. Capitol Records, LLC v. Vimeo LLC , 972 F. Supp.2d 500, 536–37 (S.D.N.Y. 2013). The portion of the Report directed to § 512(c) begins by stating that the Office “sees no reason—and none has been offered—why the section 512 ‘safe harbor’ from liability ... should not apply to the use of pre-1972 sound recordings.” Id. at 130. It observes that § 512 was “innovative legislation” and that “the concept of providing safe harbors for certain good faith acts on the Internet remains a sound principle.” Id. The Report found “no policy justification to exclude older sound recordings from section 512.” Id. We agree completely with those conclusions.

Nonetheless, the Report concluded that § 512(c)'s safe harbor does not apply to pre-1972 sound recordings, which are protected only by state law. Id. at 132. The Report rejected interpreting § 301(c) as prohibiting “all subsequent regulation [by Congress] of pre-1972 recordings,” id. at 131, but nonetheless concluded that “Congress did [not] in fact subsequently regulate pre-1972 sound recordings in section 512(c).” Id.

The Pre-1972 Sound Recordings Report arrived at its conclusion that § 512(c)'s safe harbor applies only to post-1972 sound recordings by the following reasoning: The term “infringement of copyright,” which is employed in § 512(c), “is defined in section 501(a) as the violation of ‘any of the exclusive rights of the copyright owner as provided by sections 106 through 122.’ ” Id. at 131. Therefore, that term, when used in § 512(c), “only refers to infringement of rights protected under title 17, and does not include infringement of rights protected under [state] law.” Id. at 131-32. The Report buttressed its conclusion by reference to two canons of statutory interpretation: (1) that exemptions from liability “must be construed narrowly, and any doubts must be resolved against the ... exemption”; and (2) that one section of a statute “cannot be interpreted in a manner that implicitly repeals another section.” Id. at 132.

While we unhesitatingly acknowledge the Copyright Office's superior expertise on the Copyright Act, we cannot accept its reading of § 512(c). It is based in major part on a misreading of the statute. The Report's main argument—that § 501(a) defines the words “infringement of copyright” as meaning infringement of the rights granted by the federal statute—misreads this provision. And as for the arguments based on canons of statutory construction, a subject not within the special expertise of the Copyright Office, we respectfully conclude that the pertinent canons were misunderstood and misapplied.

The Report begins its analysis by asserting that § 512(c)'s term “infringement of copyright” is defined in § 501(a) as the violation of “any of the exclusive rights of the copyright owner as provided by sections 106 through 122.” Section 501(a), however, does not contain such a definition. The Copyright Act's definitions are set forth in § 101, and do not include a definition for “infringement of copyright.” William F. Patry, Patry on Copyright § 9:1 (2016) (“The Copyright Act does not define ‘infringement.’ ”) Neither does § 501(a) purport to define “infringement of copyright.” It reads: “Anyone who violates any of the exclusive rights of the copyright owner as provided by sections 106 through 122 ... is an infringer of the copyright.” The statement that one who violates rights identified in specified sections is an “infringer of copyright” does not purport to set forth an exclusive definition of “infringer of copyright.” This provision of § 501(a) is in no way incompatible with interpreting the safe harbor as applying to infringement of state copyright laws. To state that conduct x violates a law is not the same thing as saying that conduct x is the only conduct that violates the law. And, in fact, within the terms of the Copyright Act, infringements are specified that are not among those specified in sections 106-122. See, e.g. , § 1309 (re: infringement of vessel hull designs).

A literal and natural reading of the text of § 512(c) leads to the conclusion that its use of the phrase “infringement of copyright” does include infringement of state laws of copyright. One who has been found liable for infringement of copyright under state laws has indisputably been found “liable for infringement of copyright.” In this instance, Congress did not qualify the phrase “infringement of copyright” by adding, as it did in other circumstances, the words, “under this title.” See, e.g. , § 106 (“Subject to sections 107 through 122, the owner of copyright under this title has the exclusive rights to do and to authorize any of the following ....); § 201(a) (“Copyright in a work protected under this title vests initially in the author or authors of the work.”). To interpret § 512(c)'s guarantee that service providers “shall not be liable ... for infringement of copyright” to mean that they may nonetheless be liable for infringement of copyright under state laws would be, at the very least, a strained interpretation—one that could be justified only by concluding that Congress must have meant something different from what it said.

In contrast, there is every reason to believe that Congress meant exactly what it said. As explained above, what Congress intended in passing § 512(c) was to strike a compromise under which, in return for the obligation to take down infringing works promptly on receipt of notice of infringement from the owner, Internet service providers would be relieved of liability for user-posted infringements of which they were unaware, as well as of the obligation to scour matter posted on their services to ensure against copyright infringement. The purpose of the compromise was to make economically feasible the provision of valuable Internet services while expanding protections of the interests of copyright owners through the new notice-and-takedown provision. To construe § 512(c) as leaving service providers subject to liability under state copyright laws for postings by users of infringements of which the service providers were unaware would defeat the very purpose Congress sought to achieve in passing the statute. Service providers would be compelled either to incur heavy costs of monitoring every posting to be sure it did not contain infringing pre-1972 recordings, or incurring potentially crushing liabilities under state copyright laws. It is not as if pre-1972 sound recordings were sufficiently outdated as to render the potential liabilities insignificant. Some of the most popular recorded music of all time was recorded before 1972, including work of The Beatles, The Supremes, Elvis Presley, Aretha Franklin, Barbra Streisand, and Marvin Gaye.

Whether we confine our examination to the plain meaning of the text, or consider in addition the purpose the text was intended to achieve, we find no reason to doubt that § 512(c) protects service providers from all liability for infringement of all copyrights established under the laws of the United States, regardless whether established by federal law or by local law under the sufferance of Congress, and not merely from liability under the federal statute.

Nor do we find persuasive force in the Report's reliance on canons of statutory interpretation. The Report argued that interpreting § 512(c) as protecting service providers from liability under state law would ignore the “general rule of statutory construction that exemptions from liability ... must be construed narrowly, and any doubts must be resolved against the one asserting the exemption.” Pre-1972 Sound Recordings Report, at 132. As authority for this “rule,” the Report cited our decision in Tasini v. New York Times Co ., 206 F.3d 161, 168 (2d Cir. 2000), aff'd , 533 U.S. 483, 121 S.Ct. 2381, 150 L.Ed.2d 500 (2001). The argument is flawed in several respects.

First, the Report's conception that, under the canon it cited, statutes “must be construed” in a certain way misconceives what such canons are. They are not rules , but rather suggestive “guides.” See Chickasaw Nation v. United States, 534 U.S. 84, 94, 122 S.Ct. 528, 151 L.Ed.2d 474 (2001) (emphasizing that “canons are not mandatory rules” and that “other circumstances evidencing congressional intent can overcome their force”). Such guides are based on commonsense logic and can aid in the interpretation of a legislature's intentions in the face of an ambiguous provision, but only to the extent that the logical propositions on which they are based make sense in the particular circumstance.

Second, the proposition cited by the Report with citation to our Tasini decision was not the proposition we espoused in Tasini. What we said in that case was that reading a statutory exception to a general rule “as broadly as appellees suggest would cause the exception to swallow the rule,” contravening the principle stated by the Supreme Court in Commissioner v. Clark , 489 U.S. 726, 739, 109 S.Ct. 1455, 103 L.Ed.2d 753 (1989), that “when a statute sets forth exceptions to a general rule, we generally construe the exceptions ‘narrowly in order to preserve the primary operation of the [provision].’ ” Tasini , 206 F.3d at 168 (quoting Clark , 489 U.S. at 739, 109 S.Ct. 1455 ). The difference between the proposition cited in the Report and the statements in Tasini and Clark is significant. The proposition of Tasini and Clark is supported by commonsense logic. When a statute sets forth a general principle, coupled with an exception to it, it is logical to assume, in the face of ambiguity in the exception, that the legislature did not intend the exception to be so broad as to leave nothing of the general principle. In contrast, the proposition stated by the Report—that exceptions must in all circumstances be construed narrowly, “and any doubts must be resolved against the one asserting the exception”—is arbitrary and without logical foundation. There is simply no reason to assume as a general proposition that a legislature intended all exceptions to all general principles to be construed narrowly—or broadly for that matter. The logical principle noted in Tasini and Clark has no application to the relationship between the general rule of § 301(c) and the exception provided by § 512(c). To construe the safe harbor of § 512(c) as protecting Internet service providers against liability under state law for posted infringements of which they were unaware establishes a tiny exception to the general principle of § 301(c) —that state law will continue for 95 years to govern pre-1972 sound recordings, without interference from the federal statute. The exception does not come close to ifying the general rule, and the principle of interpretation cited in Tasini therefore has no application to these facts. Further, the proposition cited by the Report is particularly without logical force where, as here, the limitation is asserted by a federal statute curtailing the operation of state law on a matter placed by the Constitution within the authority of Congress.

We also disagree with the Report's citation of Tennessee Valley Authority v Hill , 437 U.S. 153, 189, 98 S.Ct. 2279, 57 L.Ed.2d 117 (1978), for the proposition that “one section of a statute cannot be interpreted in a manner that implicitly repeals another section.” Pre-1972 Sound Recordings Report, at 132. The Report substantially overstated, and misapplied, what the Supreme Court said, which was merely that “repeals by implication are not favored.” Hill , 437 U.S. at 189, 98 S.Ct. 2279 (quoting Morton v. Mancari , 417 U.S. 535, 549, 94 S.Ct. 2474, 41 L.Ed.2d 290 (1974) ). The argument rejected by the Supreme Court was that Congress, by repeatedly funding the construction of a dam, had by implication repealed a provision of federal law protecting a wildlife species, the snail darter, whose habitat would be harmed by the operation of the dam. Id. Those circumstances had little in common with this one. Here, to the extent that Congress can be said to have repealed by § 512(c) an aspect of the rule it had previously enacted in § 301(c), it was not by implication but by specific statement. In the Hill case, the appropriations funding the dam had made no mention of any rule affecting protection of the snail darter, so that repeal through those acts of appropriation could only have been by implication . Here, in contrast, the partial repeal of § 301(c) was by the explicit statement in § 512(c) that “[a] service provider shall not be liable ... for infringement of copyright ....” The Hill principle has no application to this issue.

Finally, construing the safe harbor of § 512(c) as not granting protection to service providers from liability for state-law-based copyright infringements would substantially defeat the statute's purposes. Internet service providers that allow the public to post works on their sites would either need to incur enormous expenses to monitor all postings to ensure the absence of infringing material (contravening the provision of § 512(m) excusing them from such obligation), or would incur state-law-based liabilities for copyright infringement by reason of user-posted infringements of which they were unaware. The financial burdens in either case would be substantial and would likely either dissuade service providers from making large investments in the expansion of the growth and speed of the Internet (which Congress sought to encourage) or perhaps cause them to charge so much for the service as to undermine substantially the public usefulness of the service Congress undertook to promote.

Although an opinion expressed by the Copyright Office in such a report does not receive Chevron deference of the sort accorded to rulemaking by authorized agencies, we do recognize the Copyright Office's intimate familiarity with the copyright statute and would certainly give appropriate deference to its reasonably persuasive interpretations of the Copyright Act. See Skidmore v Swift & Co ., 323 U.S. 134, 140, 65 S.Ct. 161, 89 L.Ed. 124 (1944) (explaining that the weight of such an interpretation “will depend upon the thoroughness evident in its consideration, the validity of its reasoning, its consistency with earlier and later pronouncements, and all those facts which give it power to persuade, if lacking power to control”). In this instance, however, for the reasons explained above, we cannot accept its interpretation of § 512(c). See PATRY at § 17:102 (stating that courts should defer to the Copyright Office's interpretation of the statute only to the extent they find it persuasive); see also Cartoon Network LP, LLLP v. CSC Holdings, Inc. , 536 F.3d 121, 129 (2d Cir. 2008) (assuming that a 2001 report by the Copyright Office “deserve[d] only Skidmore deference, deference based on its ‘power to persuade,’ ” and rejecting the Office's interpretation as unpersuasive). We conclude that the safe harbor established by § 512(c) protects a qualifying service provider from liability for infringement of copyright under state law. We therefore vacate the district court's grant of summary judgment to Plaintiffs as to the availability of the DMCA safe harbor to Vimeo in relation to liability for infringement of pre-1972 sound recordings.

II. Red Flag Knowledge of Infringement

The second certified question is “Whether, under Viacom Int'l, Inc. v. YouTube, Inc., a service provider's viewing of a user-generated video containing all or virtually all of a recognizable, copyrighted song may establish ‘facts and circumstances' giving rise to ‘red flag’ knowledge of infringement” within the meaning of § 512(c)(1)(A)(ii). We consider this question in relation to the district court's denial of Vimeo's motion for summary judgment on a number of videos that conform to the facts specified in the district court's question. The district court's formulation of the question in connection with its ruling suggests that the court based its denial on the presence of the facts specified in the question. We conclude that Plaintiffs' establishment of those facts is insufficient to prove red flag knowledge. We therefore vacate the court's order denying Vimeo summary judgment as to red flag knowledge with respect to those videos.

Our court explained in Viacom that, in order to be disqualified from the benefits of the safe harbor by reason of red flag knowledge under § 512(c)(1)(A)(ii), the service provider must have actually known facts that would make the specific infringement claimed objectively obvious to a reasonable person.

The difference between actual and red flag knowledge is ... not between specific and generalized knowledge, but instead between a subjective and an objective standard. In other words, the actual knowledge provision turns on whether the provider actually or ‘subjectively’ knew of specific infringement, while the red flag provision turns on whether the provider was subjectively aware of facts that would have made the specific infringement ‘objectively’ obvious to a reasonable person.

Viacom , 676 F.3d at 31.

The hypothetical “reasonable person” to whom infringement must be obvious is an ordinary person—not endowed with specialized knowledge or expertise concerning music or the laws of copyright. Furthermore, as noted above, § 512(m) makes clear that the service provider's personnel are under no duty to “affirmatively seek[ ]” indications of infringement. The mere fact that an employee of the service provider has viewed a video posted by a user (absent specific information regarding how much of the video the employee saw or the reason for which it was viewed), and that the video contains all or nearly all of a copyrighted song that is “recognizable,” would be insufficient for many reasons to make infringement obvious to an ordinary reasonable person, who is not an expert in music or the law of copyright. Because the district court's denial of Vimeo's motion for summary judgment and concomitant certification of this question suggest that the district court believed that the evidence described in the question, without more, could render the service provider ineligible for the safe harbor, and relied on this proposition to deny summary judgment in every instance in which there was evidence that an employee of Vimeo had seen at least a portion of a video that contained substantially all of a “recognizable” copyrighted song, we vacate the district court's ruling on this question and remand for reconsideration in light of our further discussion of the standard for red flag knowledge.

A significant aspect of our ruling relates to the burdens of proof on the question of the defendant's entitlement to the safe harbor—particularly with respect to the issue of red flag knowledge. The issue is potentially confusing because of the large numbers of factual questions that can arise in connection with a claim of the safe harbor. A service provider's entitlement to the safe harbor is properly seen as an affirmative defense, and therefore must be raised by the defendant. The defendant undoubtedly bears the burden of raising entitlement to the safe harbor and of demonstrating that it has the status of service provider, as defined, and has taken the steps necessary for eligibility. On the other hand, on the question whether the service provider should be disqualified based on the copyright owner's accusations of misconduct—i.e., by reason of the service provider's failure to act as the statute requires after receiving the copyright owner's notification or otherwise acquiring actual or red flag knowledge—the burden of proof more appropriately shifts to the plaintiff. The service provider cannot reasonably be expected to prove broad negatives, providing affidavits of every person who was in its employ during the time the video was on its site, attesting that they did not know of the infringement and did not know of the innumerable facts that might make infringement obvious. And to read the statute as requiring a trial whenever the plaintiff contests the credibility of such attestations would largely destroy the benefit of the safe harbor Congress intended to create.

The Nimmer copyright treatise, noting Congress's failure to prescribe a roadmap, and observing that “courts [must] muddle through,” furnishes valuable guidance on the shifting allocation of burdens of proof as to a service provider's entitlement to the protection of the safe harbor. See Melville B. Nimmer & David Nimmer, Nimmer on Copyright § 12B.04[A][1][d], n.145 (2015). According to Nimmer, the service provider initially establishes entitlement to the safe harbor by showing that it meets the statutory definition of an eligible service provider (on whose website the allegedly infringing material was placed by a user), and that it has taken the general precautionary steps against infringement that are specified in the statute. The service provider could nonetheless be denied the safe harbor if the plaintiff-rightsholder showed that the service provider had actual knowledge, or red flag knowledge, of the infringement. The burden of proof with respect to actual or red flag knowledge would be on the plaintiff.

Acknowledging that the burden lies on the defendant to establish the affirmative defense, Nimmer explains,

It would seem that defendant may do so [establish entitlement to the safe harbor] by demonstrating that it qualifies as a service provider under the statutory definition, which has established a repeat infringer policy and follows the requisite technical measures. In terms of mental state, the burden would then appear to shift back to plaintiff. To disqualify defendant from the safe harbor, the copyright claimant must show defendant's actual knowledge or a ‘red flag’ waving in its face. But defendant can still qualify for the safe harbor if, after gaining the requisite mental state, it acted expeditiously to disable access to the infringing content. As to that last matter [expeditious take-down], the burden would seem to rest on defendant.

Id. (internal citations omitted)(emphasis added).

We agree with Nimmer's proposed allocation of shifting burdens of proof. Proper allocation of the burden of proof will necessarily have an important bearing on determining entitlements to summary judgment. Following Nimmer's cogent analysis, it appears that a defendant would, in the first instance, show entitlement to the safe harbor defense by demonstrating its status as a service provider that stores users' material on its system, that the allegedly infringing matter was placed on its system by a user, and that it has performed precautionary, protective tasks required by § 512 as conditions of eligibility, including that it adopted and reasonably implemented a policy designed to exclude users who repeatedly infringe, that it designated an agent for receipt of notices of infringement, and that it accommodates standard technical measures used by copyright owners to detect infringements.

On the issue of disqualifying knowledge, however, the burden falls on the copyright owner to demonstrate that the service provider acquired knowledge of the infringement, or of facts and circumstances from which infringing activity was obvious, and failed to promptly take down the infringing matter, thus forfeiting its right to the safe harbor. The plaintiff is, of course, entitled to take discovery of the service provider to enable it to make this showing. A copyright owner's mere showing that a video posted by a user on the service provider's site includes substantially all of a recording of recognizable copyrighted music, and that an employee of the service provider saw at least some part of the user's material, is insufficient to sustain the copyright owner's burden of proving that the service provider had either actual or red flag knowledge of the infringement. That is so for many reasons.

First, the employee's viewing might have been brief. The fact that an employee viewed enough of a video to post a brief comment, add it to a channel (such as kitten videos ) or hit the “like” button, would not show that she had ascertained that its audio track contains all or virtually all of a piece of music.

Second, the insufficiency of some viewing by a service provider's employee to prove the viewer's awareness that a video contains all or virtually all of a song is all the more true in contemplation of the many different business purposes for which the employee might have viewed the video. The purpose of the viewing might include application of technical elements of computer expertise, classification by subject matter, sampling to detect inappropriate obscenity or bigotry, and innumerable other objectives having nothing to do with recognition of infringing music in the soundtrack. Furthermore, the fact that music is “recognizable” (which, in its dictionary definition of “capable of being recognized” would seem to apply to all music that is original and thus distinguishable from other music), or even famous (which is perhaps what the district court meant by “recognizable”), is insufficient to demonstrate that the music was in fact recognized by a hypothetical ordinary individual who has no specialized knowledge of the field of music. Some ordinary people know little or nothing of music. Lovers of one style or category of music may have no familiarity with other categories. For example, 60-year-olds, 40-year-olds, and 20-year-olds, even those who are music lovers, may know and love entirely different bodies of music, so that music intimately familiar to some may be entirely unfamiliar to others.

Furthermore, employees of service providers cannot be assumed to have expertise in the laws of copyright. Even assuming awareness that a user posting contains copyrighted music, the service provider's employee cannot be expected to know how to distinguish, for example, between infringements and parodies that may qualify as fair use. Nor can every employee of a service provider be automatically expected to know how likely or unlikely it may be that the user who posted the material had authorization to use the copyrighted music. Even an employee who was a copyright expert cannot be expected to know when use of a copyrighted song has been licensed. Additionally, the service provider is under no legal obligation to have its employees investigate to determine the answers to these questions.

It is of course entirely possible that an employee of the service provider who viewed a video did have expertise or knowledge with respect to the market for music and the laws of copyright. The employee may well have known that the work was infringing, or known facts that made this obvious. The copyright owner is entitled to discovery in order to obtain the specific evidence it needs to sustain its burden of showing that the service provider did in fact know of the infringement or of facts that made infringement obvious. But the mere fact that a video contains all or substantially all of a piece of recognizable, or even famous, copyrighted music and was to some extent viewed (or even viewed in its entirety) by some employee of a service provider would be insufficient (without more) to sustain the copyright owner's burden of showing red flag knowledge.

Plaintiff argues that, under this interpretation of the standard for finding red flag knowledge, red flag knowledge is so similar to actual knowledge of infringement as to violate the rule of statutory interpretation that no portion of the statute should be interpreted in a manner that renders it superfluous. This argument has no merit. While the difference between actual knowledge of infringement under § 512(c)(1)(A)(i) and red flag knowledge under § 512(c)(1)(A)(ii) may not be vast, it is nonetheless a real difference. If the facts actually known by an employee of the service provider make infringement obvious, the service provider cannot escape liability through the mechanism of the safe harbor on the ground that the person with knowledge of those facts never thought of the obvious significance of what she knew in relation to infringement. Plaintiffs further argue that this understanding of red flag knowledge reduces it to a very small category. Assuming this is so, it is of no significance. The fact that Congress was unwilling to extend the safe harbor to circumstances where the service provider did not subjectively know that the posted material infringed, but did know facts that made infringement objectively obvious, does not compel the conclusion that Congress expected this extension to cover a large number of instances. That is especially so in view of the fact that the purpose of § 512(c) was to give service providers immunity, in exchange for augmenting the arsenal of copyright owners by creating the notice-and-takedown mechanism.

In sum, a showing by plaintiffs of no more than that some employee of Vimeo had some contact with a user-posted video that played all, or nearly all, of a recognizable song is not sufficient to satisfy plaintiffs' burden of proof that Vimeo forfeited the safe harbor by reason of red flag knowledge with respect to that video. As it appears that the district court employed that inappropriate standard as the basis for its denial of Vimeo's motion for summary judgment on numerous videos conforming to that description, we vacate those rulings and remand for further consideration. Vimeo is entitled to summary judgment on those videos as to the red flag knowledge issue, unless plaintiffs can point to evidence sufficient to carry their burden of proving that Vimeo personnel either knew the video was infringing or knew facts making that conclusion obvious to an ordinary person who had no specialized knowledge of music or the laws of copyright.

III. Willful Blindness

Our final issue on this appeal involves Plaintiffs' contention that the district court, in rejecting their claim of willful blindness, misapplied our teachings in Viacom, which recognized that “the willful blindness doctrine may be applied, in appropriate circumstances, to demonstrate knowledge or awareness of specific instances of infringement under the DMCA.” Viacom , 676 F.3d at 35. We disagree with Plaintiffs' argument and see no reason to disturb the district court's ruling.

Plaintiffs essentially make three arguments. First, based on evidence that Vimeo monitored videos for infringement of visual content but not for infringement of audio content, they argue that they have demonstrated willful blindness to infringement of music, which justifies liability under Viacom. Their second argument is that Vimeo's awareness of facts suggesting a likelihood of infringement gave rise to a duty to investigate further, and that Vimeo's failure to do so showed willful blindness that justifies liability. Finally, they argue that, having encouraged users to post infringing matter, Vimeo could not then close its eyes to the resulting infringements without liability.

The first two arguments are easily disposed of. As we made clear in Viacom, § 512(m) relieves the service provider of obligation to monitor for infringements posted by users on its website. We see no reason why Vimeo's voluntary undertaking to monitor videos for infringement of visual material should deprive it of the statutory privilege not to monitor for infringement of music. Plaintiffs' argument is refuted by § 512(m).

Their second argument, that awareness of facts suggesting a likelihood of infringement gave rise to a duty to investigate further, does not fare better. Section 512(c) specifies the consequences of a service provider's knowledge of facts that might show infringement. If the service provider knows of the infringement, or learns of facts and circumstances that make infringement obvious, it must act expeditiously to take down the infringing matter, or lose the protection of the safe harbor. But we can see no reason to construe the statute as vitiating the protection of § 512(m) and requiring investigation merely because the service provider learns facts raising a suspicion of infringement (as opposed to facts making infringement obvious ). Protecting service providers from the expense of monitoring was an important part of the compromise embodied in the safe harbor. Congress's objective was to serve the public interest by encouraging Internet service providers to make expensive investments in the expansion of the speed and capacity of the Internet by relieving them of burdensome expenses and liabilities to copyright owners, while granting to the latter compensating protections in the service providers' takedown obligations. If service providers were compelled constantly to take stock of all information their employees may have acquired that might suggest the presence of infringements in user postings, and to undertake monitoring investigations whenever some level of suspicion was surpassed, these obligations would largely undo the value of § 512(m). We see no merit in this argument.

Plaintiffs' third argument may fare better in theory, but is not supported by the facts of this case, at least as we understand them. In Viacom, we made clear that actual and red flag knowledge under the DMCA ordinarily must relate to “specific infringing material,” id. at 30, and that, because willful blindness is a proxy for knowledge, id. at 34–35, it too must relate to specific infringements. Plaintiffs argue, however, that Vimeo, in order to expand its business, actively encouraged users to post videos containing infringing material. They argue that, notwithstanding the formulation in Viacom, a service provider cannot adopt a general policy of urging or encouraging users to post infringing material and then escape liability by hiding behind a disingenuous claim of ignorance of the users' infringements.

We need not decide whether Plaintiffs' proposed gloss on Viacom is correct as a matter of law. Assuming that it is, Plaintiffs still cannot rely on such a theory in this instance. The evidence cited to us by Plaintiffs, consisting of a handful of sporadic instances (amongst the millions of posted videos) in which Vimeo employees inappropriately encouraged users to post videos that infringed music cannot support a finding of the sort of generalized encouragement of infringement supposed by their legal theory. It therefore cannot suffice to justify stripping Vimeo completely of the protection of § 512(m). Moreover, because that evidence was not shown to relate to any of the videos at issue in this suit, it is insufficient to justify a finding of red flag knowledge, under the principle of Viacom, as to those specific videos. Thus, notwithstanding a few unrelated instances in which its employees improperly encouraged specific infringements, Vimeo can still assert the protection of § 512(m) for the present suit, and claim the benefit of the safe harbor, in the absence of a showing by Plaintiffs of facts sufficient to demonstrate that Vimeo, having actual or red flag knowledge of infringement in the videos that are the subject of Plaintiffs' suit, failed to promptly take them down.

CONCLUSION

We conclude: (1) The safe harbor of § 512(c) of the DMCA does apply to pre-1972 sound recordings, and therefore protects service providers against liability for copyright infringement under state law with respect to pre-1972 sound recordings, as well as under the federal copyright law for post-1972 recordings. The district court's grant of partial summary judgment to Plaintiffs with respect to Vimeo's entitlement to the safe harbor for infringements of pre-1972 recordings is therefore vacated. (2) The various factual issues that arise in connection with a service provider's claim of the safe harbor of § 512(c) are subject to shifting burdens of proof, as described above. Because, on a defendant's claim of the safe harbor of § 512(c), the burden of showing facts supporting a finding of red flag knowledge shifts to the plaintiff, and the district court appears to have denied Vimeo's motion for summary judgment as to a number of videos on this issue based on a test that would improperly deny service providers access to the safe harbor, we vacate the court's denial of Vimeo's motion for summary judgment on that issue, and remand for reconsideration and further proceedings. (3) We reject Plaintiffs' argument that the district court erred in its ruling in Vimeo's favor as to the Plaintiffs' reliance on the doctrine of willful blindness.

The district court's rulings are accordingly affirmed in part and vacated in part and the matter is remanded for further proceedings.