ChatGPT Lawsuits

Plaintiff. The New York Times Company. Filed December 27, 2023 in the Southern District of New York. Microsoft is a co-defendant alongside OpenAI because the complaint reaches the use of OpenAI models inside Microsoft's Bing Chat / Copilot products as well as inside ChatGPT.

Theory of liability. Two intertwined claims. First, that OpenAI copied millions of Times articles into the training corpus without licensing them — the same training-input theory at issue in Bartz v. Anthropic on the books side. Second, that the resulting models reproduce paywalled Times content in outputs and are marketed as substitutes for a Times subscription — a market-substitution claim that goes beyond a pure ingestion theory. Plaintiffs added DMCA Section 1202 (copyright-management-information) claims and a tortious-interference claim alongside the direct copyright counts.

Procedural posture. A motion-to-dismiss ruling in 2024 narrowed the case but did not gut it — the direct copyright counts and the DMCA Section 1202 count survived; some adjacent claims fell out. Discovery has produced contested fights over the training-corpus contents and the methodology behind the Times's output-reproduction examples (the so-called “hallucinated paywall bypass” demonstrations attached to the complaint). The case has consolidated procedurally with the broader S.D.N.Y. publisher docket but retains the Times as named flagship plaintiff.

Why it matters. This is the single most-watched fair-use case against a generative-AI defendant. Where Bartz resolved the books-side training question on summary judgment via the Alsup ruling (training fair use; piracy not), NYT v. OpenAI tests the same line in a different posture — news articles instead of books, market substitution as a theory, and the Second Circuit instead of the Ninth.

What to watch next. Summary judgment is the inflection point, when the court has to apply fair use against a developed factual record on training inputs and output behavior. Whether the court adopts or distinguishes the Alsup framing will reset the precedent map.

Plaintiffs. The Authors Guild and a class of book authors including George R. R. Martin, John Grisham, Jodi Picoult, David Baldacci, Michael Connelly, Jonathan Franzen, Elin Hilderbrand, and George Saunders. Filed September 2023 in the Southern District of New York. The case has been consolidated with the parallel author class actions originally filed in the Northern District of California (Tremblay, Silverman, Awad, Chabon) for management purposes.

Theory of liability. The same training-input copyright theory at issue in Bartz v. Anthropic on the books side: that OpenAI copied entire copyrighted books into the training corpus to train GPT-3, GPT-3.5, and GPT-4, infringing the authors' exclusive reproduction right. The Northern California branch of the litigation (Tremblay, Silverman, Awad, Chabon) added DMCA Section 1202, negligence, and unjust enrichment counts; most of those non-direct-infringement claims were dismissed in early-2024 motion practice. The direct copyright claims survive.

Procedural posture. Active in S.D.N.Y. after the procedural consolidation. Class-certification briefing and discovery on the corpus contents are the dominant procedural action. Settlement discussions have been reported intermittently in trade-press coverage but no settlement has been entered.

Why it matters. This is the OpenAI-side equivalent of Bartz: the same plaintiff archetype (named book authors), the same theory (training on copyrighted books without license), the same defense (training is transformative fair use). The Anthropic case settled before trial; whether the OpenAI case follows that path or pushes through to a substantive fair-use ruling will set the second federal-trial-court line on the question.

What to watch next. Class certification, summary judgment, and the read-across of the Alsup ruling. Counsel on both sides have publicly cited Bartz in filings.

Cases in the cluster.

• Daily News, Tribune, MediaNews Group, et al. v. OpenAI & Microsoft (S.D.N.Y., April 2024). A coalition of eight regional newspaper publishers including the New York Daily News, the Chicago Tribune, the Denver Post, the Orlando Sentinel, the San Jose Mercury News, the Orange County Register, the St. Paul Pioneer Press, and the Mercury News. Owned in part by Alden Global Capital. Same training-corpus and output-reproduction theory as NYT.
• Center for Investigative Reporting (Mother Jones / Reveal) v. OpenAI & Microsoft (S.D.N.Y., June 2024). The first nonprofit-news plaintiff in the docket. CIR alleges its investigative reporting was ingested into the training corpus without license and that ChatGPT reproduces it on demand without attribution.
• The Intercept Media v. OpenAI & Microsoft (S.D.N.Y., February 2024). Pleads DMCA Section 1202 violations — that copyright-management information (author bylines, publication dates) was stripped from Intercept articles during training. The DMCA-1202 angle differs from the direct-infringement framing the other publisher cases lead with.
• Raw Story Media & AlterNet Media v. OpenAI & Microsoft (S.D.N.Y., February 2024). Companion DMCA Section 1202 case to The Intercept. The S.D.N.Y. court dismissed the DMCA Section 1202 claims for lack of Article III standing in late 2024; plaintiffs pursued an amended complaint and have appealed the standing ruling to the Second Circuit.

Theory of liability. Two-track. The Daily News coalition and CIR run the direct-copyright theory: training-corpus ingestion plus output reproduction, on the same logic as the Times. The Intercept and Raw Story cases are the DMCA Section 1202 frontier — the claim that removing or altering CMI during training is itself a statutory violation independent of the underlying infringement question.

Procedural posture. The cases are pending in S.D.N.Y. and have been informally coordinated for discovery purposes alongside the NYT case, given the overlapping factual questions about the training corpus and OpenAI's output behavior. The Raw Story standing dismissal is on appeal.

Why it matters. The publisher coalition tests two things the Times case alone doesn't reach: (1) whether the same training-fair-use analysis travels across publisher size, business model (for-profit chain, nonprofit, alternative weekly), and content type; and (2) whether DMCA Section 1202 is a viable independent theory against AI training. The Second Circuit's ruling on the Raw Story standing appeal is the load-bearing piece for the latter.

What to watch next. The Raw Story Second Circuit appeal on Article III standing for DMCA Section 1202 claims. After that, motion-to-dismiss outcomes in the Daily News and CIR cases that test whether the NYT motion-to-dismiss ruling reads across cleanly to other publishers.

Plaintiff. Elon Musk, individually and (in the federal refiling) as the controlling owner of xAI Corp. Musk was an OpenAI cofounder and board member from 2015 until his February 2018 departure, and had committed (per the complaint) approximately $44 million in startup funding before his exit.

Procedural history. Filed February 29, 2024 in California state court (San Francisco Superior). Voluntarily dismissed by Musk in June 2024 on the eve of a scheduled hearing on OpenAI's demurrer. Refiled August 2024 in the Northern District of California with an expanded complaint adding federal RICO claims and additional defendants (including Microsoft and several individual board members). The refiling triggered a second motion-to-dismiss round; some claims were dismissed and others survived. Musk separately sought a preliminary injunction to block OpenAI's for-profit conversion, which the court denied in early 2025 while leaving the underlying breach claims in place.

Theory of liability. Not copyright. The complaint pleads breach of contract (the “founding agreement” alleged to be embodied in the original 2015 articles of incorporation, the early board materials, and Musk's email correspondence with Altman and Brockman), breach of fiduciary duty against the cofounder defendants in their officer-and-director capacities, unjust enrichment, false advertising, unfair competition, and (in the federal refiling) racketeering and wire-fraud predicates supporting the civil RICO claim. The animating allegation is that OpenAI was founded as a nonprofit committed to safe, broadly-shared AGI and was converted into a for-profit business operating as Microsoft's contractual partner, in betrayal of the founding promise.

Why it matters. This is the dispute over OpenAI's corporate structure that every other piece of the docket touches. The November 2023 board episode, the 2024 leadership exodus, the 2024 – 2025 for-profit-conversion fight, and Microsoft's contractual position all sit downstream of the same structural question: what was OpenAI agreed to be, and who has standing to enforce that. Musk v. Altman is the case that puts the question into a federal courtroom directly. It is part legal docket and part corporate-governance episode — the broader for-profit-conversion narrative lives on the ChatGPT Leadership page alongside the November 2023 board episode; the legal-procedure half lives here.

What to watch next. Whether the surviving claims reach summary judgment or trial, whether the for-profit-conversion injunction ruling is appealed, and whether the case settles — an outcome whose plausibility moves with Musk's xAI competitive posture and OpenAI's structural plans.

Plaintiffs. Anonymous open-source software developers (proceeding as Doe 1, Doe 2, etc.) on behalf of a putative class of GitHub repository owners whose code was used to train the Codex model that powered the original GitHub Copilot. Filed November 3, 2022 in the Northern District of California. Defendants are GitHub, Microsoft, and OpenAI — OpenAI as the trainer of the Codex model GitHub deployed.

Theory of liability. The complaint pleaded a stack of theories layered on top of the core training-data copyright claim: direct copyright infringement, DMCA Section 1202 violations (CMI stripping during training), open-source license violations (MIT, Apache, GPL, BSD attribution requirements), breach of contract on the GitHub Terms of Service, unfair competition, and unjust enrichment.

Procedural posture. The court dismissed most of the layered claims early. The DMCA Section 1202 counts were dismissed for lack of plausible identicality between the trained outputs and the asserted source code. The open-source license claims were dismissed for failure to plead the elements of contract formation and breach with particularity. What survives is the core direct-copyright infringement theory, on a narrower factual basis — specifically tied to outputs that reproduce identifiable copyrighted code from named plaintiffs' repositories.

Why it matters. Doe v. GitHub Copilot was the first major lawsuit specifically targeting generative-AI training, and the early dismissals shaped the pleading map for everything that came after. Plaintiffs' counsel in subsequent cases (including Tremblay and Silverman) explicitly trimmed their DMCA Section 1202 and unjust-enrichment theories in response to the Doe dismissals, and the surviving direct-copyright frame became the dominant theory across the AI-training docket. The case also tested whether open-source licenses are reachable through training; the answer so far is “not as pleaded.”

What to watch next. Whether the surviving direct-copyright claims reach class certification with the anonymized plaintiffs, and whether Copilot's evolution beyond the original Codex model (later versions of Copilot use GPT-4 and successor models) alters the factual posture of the output-reproduction theory.