From Casetext: Smarter Legal Research

Tremblay v. Openai, Inc.

United States District Court, Northern District of California
Aug 8, 2024
23-cv-03223-AMO (N.D. Cal. Aug. 8, 2024)

Opinion

23-cv-03223-AMO

08-08-2024

PAUL TREMBLAY, et al., Plaintiffs, v. OPENAI, INC., et al., Defendants.


ORDER GRANTING PLAINTIFFS' MOTION FOR RELIEF FROM DISCOVERY ORDER RE: DKT. NO. 157

ARACELI MARTÍNEZ-OLGUÍN, United States District Judge

Pursuant to Federal Rule of Civil Procedure 72(a) and Northern District of California Civil Local Rule 72-2, Plaintiffs move for relief from Magistrate Judge Robert M. Illman's Nondispositive Pretrial Order (ECF 154 “Order”). ECF 157 (“Mot”). The Order granted Defendants' request to compel OpenAI account information, prompts, and outputs for Plaintiffs' testing of ChatGPT in connection with their pre-suit investigation. On July 15, 2024, this Court gave Defendants an opportunity to respond to Plaintiffs' motion for relief. See ECF 159. Defendants filed a response on July 24, 2024. ECF 161. Having read the parties' papers and carefully considered their arguments and the relevant legal authority, the Court hereby GRANTS Plaintiffs' motion for the following reasons.

I. BACKGROUND

Plaintiffs, on behalf of themselves and a putative class, assert a claim for direct copyright infringement against OpenAI. First Consolidated Amended Complaint (“FCAC”) (ECF 120). Plaintiffs allege that Defendants' ChatGPT is powered by software programs known as large language models, which were trained through “copying massive amounts of text and extracting expressive information from it.” FCAC ¶ 2. They allege that their copyrighted books were used to train ChatGPT and that when prompted, ChatGPT generates accurate summaries of their works. FCAC ¶¶ 4-5, 51. Plaintiffs attached to their FCAC an exhibit - Exhibit B - that has prompts used to ask ChatGPT to summarize in detail various parts of Plaintiffs' books, as well as the outputs produced by ChatGPT. FCAC ¶ 51; FAC Ex. B (ECF 12-2) at 2-40.

OpenAI served a document request on Plaintiffs (“RFP 9”), which Defendants subsequently narrowed to seek: “(a) the OpenAI account information for individuals who used ChatGPT to investigate Plaintiffs' claims; and (b) the prompts and outputs for Plaintiffs' testing of ChatGPT in connection with their pre-suit ChatGPT testing, including prompts and outputs that did not reproduce or summarize Plaintiffs' works or otherwise support Plaintiffs' claims, along with documentation of Plaintiffs' testing process.” ECF 153 at 1. Plaintiffs offered to produce “full threads of the prompts and outputs” that led to the examples attached in Exhibit B to the FCAC. Id. However, Plaintiffs refused to produce prompts and outputs that were not used in the FCAC, and which did not reproduce or summarize Plaintiffs' works or support their claims. Id. at 2-3.

Plaintiffs and Defendants filed a joint discovery statement as required by Magistrate Judge Illman's standing order. Judge Illman granted Defendants' request to compel (1) OpenAI account information for individuals who used ChatGPT to investigate Plaintiffs' claims; (2) the prompts and outputs for Plaintiffs' testing of ChatGPT in connection with their pre-suit testing, including prompts and outputs that did not reproduce, summarize, or support Plaintiffs' claims; and (3) documentation of Plaintiffs' testing process. ECF 154 at 7. The instant motion followed.

II. LEGAL STANDARD

The standard for review of a magistrate judge's non-dispositive ruling under Federal Rule of Civil Procedure 72(a) is highly deferential. “The reviewing court may not simply substitute its judgment for that of the deciding court.” Grimes v. City & Cnty. of S.F., 951 F.2d 236, 241 (9th Cir. 1991). Instead, “[a] non-dispositive order entered by a magistrate must be deferred to unless it is ‘clearly erroneous or contrary to law.'” Id. (quoting Fed.R.Civ.P. 72(a)). “There is clear error only when the court is ‘left with the definite and firm conviction that a mistake has been committed.'” Zepeda v. Paypal, Inc., 2014 WL 4354386, at *3 (N.D. Cal. Sept. 2, 2014) (quoting Easley v. Cromartie, 532 U.S. 234, 242 (2001)). “A decision may be contrary to law if it fails to apply or misapplies relevant statutes, case law, or rules of procedure.” Rojas v. Bosch Solar Energy Corp., No. 18-CV-05841-BLF, 2020 WL 6557547, at *3 (N.D. Cal. July 20, 2020) (citation omitted).

III. DISCUSSION

The Order grants Defendants' request to compel, finding that the account settings and negative test results are fact work product, that Plaintiffs waived the ability to assert work product protection by placing a large subset of those facts in the FCAC and Exhibit B, and that an evaluation of the negative results and account settings is necessary to Defendants' ability to understand Plaintiffs' positive results. ECF 154 at 5-6. Plaintiffs challenge each of these findings as clear error or contrary to law.

Federal Rule of Civil Procedure 26 protects disclosure of documents and tangible things “prepared in anticipation of litigation or for trial.” Fed.R.Civ.P. 26(b)(3)(A); see In re Grand Jury Subpoena (Mark Torf/Torf Env't Mgmt.), 357 F.3d 900, 906 (9th Cir. 2004). Unlike fact work product, “opinion work product” includes “an attorney's mental impressions, conclusions, opinions, or legal theories developed in anticipation of litigation.” Republic of Ecuador v. Mackay, 742 F.3d 860, 869 n.3 (9th Cir. 2014); see Fed.R.Civ.P. 26(b)(3)(B). It “is virtually undiscoverable.” Mackay, 742 F.3d at 869 n.3 (citations omitted).

The Order found that the ChatGPT account settings and negative test results are “more in the nature of fact work product than opinion work product” because they are “more in the nature of bare facts.” ECF 154 at 5-6. This is a misapplication of law as the ChatGPT prompts were queries crafted by counsel and contain counsel's mental impressions and opinions about how to interrogate ChatGPT, in an effort to vindicate Plaintiffs' copyrights against the alleged infringements. See Mackay, 742 F.3d at 869 n.3.

Plaintiffs also take issue with the Order's broad finding of waiver as contrary to law. While the central purpose of the work-product doctrine is to “prevent exploitation of a party's efforts in preparing for litigation,” it is a “qualified” privilege that can be waived by a litigant by “reveal[ing] or plac[ing] the work product at issue during the course of litigation.” United States v. Sanmina Corp., 968 F.3d 1107, 1119 (9th Cir. 2020) (citations omitted). “[C]ourts differentiate between a waiver of factual work product and a waiver of opinion work product.” Munguia-Brown v. Equity Residential, 337 F.R.D. 509, 516 (N.D. Cal. 2021). Waiver of fact work product extends to the subject matter of the disclosed work product. See Hernandez v. Tanninen, 604 F.3d 1095, 1100 (9th Cir. 2010) (concluding that disclosure of attorney notes constituted a waiver over the subject matter of the disclosed notes). The scope of any waiver “must be closely tailored . . . to the needs of the opposing party” and courts must “be careful to impose a waiver no broader than needed to ensure the fairness of the proceedings before it.” Sanmina, 968 F.3d at 1122, 1124. However, opinion work product is discoverable by waiver only where “mental impressions are at issue in a case and the need for the material is compelling.” Id. at 1124-25 (quoting Holmgren v. State Farm Mut. Auto. Ins. Co., 976 F.2d 573, 577 (9th Cir. 1992)); see In re Cathode Ray Tube Antitrust Litig., No. 07-CV-05944-JST, 2023 WL 5667880, at *5 (N.D. Cal. July 6, 2023).

Because the Order determined that the prompts and outputs were fact work product, the Order concluded that broad subject matter waiver applied, and that Plaintiffs had waived the privilege as to all account settings, prompts, and outputs for Plaintiffs' testing of ChatGPT. ECF 154 at 6. While a narrow subject matter waiver would apply to fact work product revealed by counsel, applying subject matter waiver here is inapposite to Ninth Circuit law related to waiver of opinion work product.

The Order also stated that subject matter waiver would apply even if the work product at issue here were opinion work product, ECF 154 at 6, however that misstates the law. See Sanmina, 968 F.3d at 1125.

The Order did not conclude - nor did Defendants show - that counsel's mental impressions are at issue here or that Defendants have a compelling need for the material. See Sanmina, 968 F.3d at 1125; see also Upjohn Co. v. United States, 449 U.S. 383, 401 (1981) (“[S]uch work product cannot be disclosed simply on a showing of substantial need and inability to obtain the equivalent without undue hardship.”). Indeed, without further support, Defendants state that the information is necessary to test Plaintiffs' allegations, ECF 153 at 3, and the Order finds that the negative test results and account settings “appear[] likely to aid in Defendants' understanding of Plaintiff's positive results ....” ECF 154 at 6. Thus, it was improper to extend the scope of the waiver to opinion work product not disclosed in the operative complaint.

IV. CONCLUSION

For the foregoing reasons, the Court GRANTS Plaintiffs' motion for relief. The Court DENIES Defendants' request to compel production of all negative testing results and the documentation of the testing process. To the extent they have not already, Plaintiffs shall disclose the prompts, outputs, and account settings used for the inputs for the positive testing results referred to and disclosed in the FCAC.

IT IS SO ORDERED.


Summaries of

Tremblay v. Openai, Inc.

United States District Court, Northern District of California
Aug 8, 2024
23-cv-03223-AMO (N.D. Cal. Aug. 8, 2024)
Case details for

Tremblay v. Openai, Inc.

Case Details

Full title:PAUL TREMBLAY, et al., Plaintiffs, v. OPENAI, INC., et al., Defendants.

Court:United States District Court, Northern District of California

Date published: Aug 8, 2024

Citations

23-cv-03223-AMO (N.D. Cal. Aug. 8, 2024)