CONCORD MUSIC GROUP, INC., et al., Plaintiffs, v. ANTHROPIC PBC, Defendant Case No. 24-cv-03811-EKL (SVK) United States District Court, N.D. California Filed November 26, 2024 van Keulen, Susan, United States Magistrate Judge ORDER RESOLVING DISCOVERY DISPUTE Re: Dkt. Nos. 245-46 Plaintiffs (the “Publishers”) accuse Defendant (“Anthropic”) of copyright infringement arising out of its Claude AI technology. In general, they assert that Anthropic developed Claude using unauthorized copies of the Publishers’ copyrighted music and that Claude reproduces unauthorized copies of that music when third parties use the technology. In connection with these claims, the Publishers served requests for production and interrogatories on Anthropic. Anthropic responded with two broad objections to the scope of the Publishers’ requests: (1) the Publishers’ definition of Claude captured irrelevant technologies; and (2) the Publishers sought documents and information originating as early as 2020, which predates the development of Claude (and Anthropic’s founding). The Parties now request that the Court rule on the validity of these threshold objections before they proceed with discovery. See Dkt. 245. They appeared for a virtual hearing on the dispute on November 21, 2024. See Dkt. 265. After considering the Parties’ joint submission, relevant law and the record in this action, and after hearing oral argument, the Court rules as follows: In light of the allegations in the complaint, prototype technologies that were: (1) “trained” with the Publishers’ copyrighted works; and (2) used to develop or integrated into Claude are within the scope of relevance under Federal Rule of Civil Procedure 26(b), even though such prototypes may not have been publicly released. Similarly, versions of Claude currently under development that Anthropic trains with the Publishers’ copyrighted works are also within the scope of relevance under Rule 26(b).[1] Accordingly, the Parties shall use the following three-part definition of Claude: With respect to technology predating the release of Claude 1, “each earlier AI model where both: (1) the Publishers’ lyrics were included in the training data for the model; and (2) the model was used to train publicly released Claude models.” With respect to publicly released versions of Claude, “Claude 1, Claude 2, Claude 2.1, the Claude 3 model family (Haiku, Sonnet and Opus) and Claude 3.5 Sonnet.” With respect to versions of Claude under development, “each AI model not captured by either of the preceding subparts of this definition where both: (1) the Publishers’ lyrics were included in the training data for the model; and (2) the model is being used to train not-yet-released Claude models.” Notwithstanding the foregoing definition, Anthropic may object to individual discovery requests, but it may do so only as appropriate in light of the Court’s admonishment below. The Court does not consider third-party technologies that integrate the Claude Application Programming Interface (“API”) to fit within the definition of Claude. Nevertheless, uses of Claude through the Claude API may be relevant. Plaintiff may serve discovery requests directed to information arising out of accessing Claude through the Claude API, and Anthropic may object to individual discovery requests as appropriate (but again, only as appropriate in light of the Court’s admonishment below). The Court declines to issue a per se ruling one way or the other on the issue of whether any information originating as early as 2020 could fall within the scope of relevant discovery. To the extent the above definition of Claude captures information originating as early as 2020, that information is not irrelevant merely because it predates Anthropic’s preferred start date of 2022. This ruling is without prejudice to Anthropic renewing its “time” objection in the future with respect to a particular discovery request as appropriate. At the November 21 hearing, Anthropic argued that permitting as broad a scope of relevancy as requested by the Publishers would place an immense burden on its discovery obligations. That concern is premature. In presenting the instant dispute, the Parties ask the Court to resolve two threshold issues of relevancy. The Court declines to rule in the abstract that a broad conception of relevancy will overburden Anthropic. Now that the preliminary issues of relevancy are resolved, Anthropic can evaluate its obligations on a request-by-request basis. Only if Anthropic determines, after conducting a good-faith evaluation, that producing requested discovery in light of the Court’s relevancy determinations would be overly burdensome can it properly object on the basis of burden. The Court admonishes the Parties to: (1) adhere closely to the requirements of Rule 26 and the undersigned’s Civil and Discovery Referral Matters Standing Order, including the requirement for robust and good-faith meet-and-confer efforts; and (2) compromise as to the scope of relevant discovery that is proportional to the needs of this litigation. Anthropic also requests that the Court seal certain portions of two documents filed in connection with the Parties’ joint submission. See Dkts. 246, 252. The Court’s rationale in resolving the instant discovery dispute does not implicate the information sought to be sealed. The Court accordingly TERMINATES AS MOOT the request to seal. Dkts. 246-2 and 246-3 shall remain under seal. SO ORDERED. Footnotes [1] As a practical matter, the Parties will have to negotiate an end date for relevant product development in accordance with the Court’s admonition contained herein.