What was filed
The complaint is Document 1 in No. 5:26-cv-12308, U.S. District Court for the Northern District of California, filed and dated October 21, 2026. It runs 27 pages. It says Exhibit A lists 853 musical compositions that Reservoir owns or exclusively controls, with their registration numbers. The plaintiffs are Reservoir Media Management, Inc., which the complaint describes as a Delaware corporation with its principal place of business in New York, New York, and two private limited companies formed under the laws of England and Wales, Reservoir Media Management (UK) Limited and Shapiro, Bernstein & Co. Limited, which the complaint says are wholly owned within the Reservoir group. The complaint calls them together "Reservoir" and describes Reservoir as an independent music company. The defendant is Anthropic PBC, described as a Delaware public benefit corporation with its principal place of business in San Francisco, California, that develops the Claude family of large language models. Counsel are Adams & Reese, LLP of Nashville, Tennessee, and Skaggs Faucette LLP of San Francisco, California, with pro hac vice admission listed as forthcoming for two attorneys. Jurisdiction is pleaded under 28 U.S.C. 1331 and 1338(a), and venue under 28 U.S.C. 1391(b) and/or 1400. The complaint lists related copyright actions against Anthropic pending in the district, including two brought by Concord Music Group, Inc. and one by BMG Rights Management (US) LLC. The plaintiffs demand a jury trial under Federal Rule of Civil Procedure 38(b). The case was assigned on October 22, 2026 to Magistrate Judge Alex G. Tse. (Compl. paras. 2, 10 to 19; pp. 1, 26 to 27; Dkt. 4.)
What the complaint alleges about training
The complaint alleges, on information and belief, that Reservoir's lyrics for each of the works were included in text Anthropic collected from the internet and in pre-assembled datasets it used to train Claude, naming Common Crawl, "Book3," and The Pile, including a YouTube Subtitles subset. It quotes a response it says Anthropic gave to requests for admission in Bartz v. Anthropic PBC, No. 4:24-cv-05417 (N.D. Cal.), that at least one commercially released Claude model "was trained on certain data derived from YouTube subtitles found in a public dataset that Anthropic understands to be a version of The Pile, or components thereof." Citing the July 17, 2025 decision in Bartz, it alleges Anthropic used BitTorrent to download millions of books from Library Genesis and other pirate libraries, including books that contain Reservoir's lyrics, and alleges, on information and belief, that torrenting also distributed copies of the works. (Compl. paras. 30 to 41.)
It alleges that Anthropic maintained, and on information and belief continues to maintain, an internal, all-purpose repository where the works are kept indefinitely apart from any training run. It alleges, on information and belief, that Anthropic's data-cleaning pipeline deliberately removed copyright management information, including song titles, songwriter and composer credits, publisher names, and copyright-ownership attributions, and that Anthropic chose an extraction tool called Newspaper over alternatives that would have preserved that information. (Compl. paras. 44 to 51.)
What the complaint alleges about outputs
The complaint alleges that Claude can reproduce Reservoir's lyrics in full or in substantial part when prompted, points to examples it says appear in Anthropic's publicly released fine-tuning dataset, and says Anthropic added output filters after the first Concord suit was filed in October 2023. It alleges, on information and belief, that those filters cover only a narrow subset of songs and are readily circumvented. It alleges that Claude Haiku 4.5 and Sonnet 5 decline to reproduce lyrics but still create derivative works, and quotes a Sonnet 5 response, which it calls an admission, given after Claude was asked to rewrite certain songs for a modern audience: "These renders keep landing very close to the originals: same structure, same hooks reused with light rewording, and full sets of verses/choruses/bridges. That's edging past 'inspired by' into reproducing the copyrighted song, even when the words are swapped out." It alleges, on information and belief, that Claude has created infringing output for all 853 works. (Compl. paras. 4, 52 to 66.)
Fair use
Anticipating a fair use defense under 17 U.S.C. 107, the complaint relies on Bartz v. Anthropic PBC, 787 F. Supp. 3d 1007 (N.D. Cal. 2025), an earlier case against Anthropic in the same district, for the point that Anthropic's practice of indefinite retention in a general-purpose archive is legally distinct from purchasing and scanning works solely to train a specific model, and says the Bartz court held that the indefinite retention of unlicensed copies is not fair use. (Compl. paras. 23, 73 to 76, 81.)
Counts and relief
Count I alleges willful direct copyright infringement under 17 U.S.C. 106 and 501, through acquisition, storage, data cleaning, and training, and through outputs that it says violate the rights in 17 U.S.C. 106(1) through 106(5). Count II alleges circumvention of technological measures that control access to the works on licensed platforms, under 17 U.S.C. 1201(a)(1). Count III alleges intentional removal or alteration of copyright management information, as defined in 17 U.S.C. 1202(c), in violation of 17 U.S.C. 1202(b)(1). The complaint seeks declarations, a permanent injunction, impoundment or destruction of copies under 17 U.S.C. 503(b), including copies in training datasets and model weights, an accounting of training data, actual damages and profits under 17 U.S.C. 504(b) or statutory damages under 17 U.S.C. 504(c) that it totals at $127,950,000, statutory damages under 17 U.S.C. 1203(c)(3)(A) and (B) that it totals at $2,132,500 and $21,325,000, attorney's fees and costs under 17 U.S.C. 505 and 1203(b), and interest. (Compl. paras. 73 to 97; pp. 25 to 26.)