AI copyright and training data laws by state
AI Laws USA tracks 29 U.S. AI rules on copyright and training data across federal, state, county, and city government. Each entry links to its official source.
Federal copyright and training data rules (23)
-
In effect
Bartz v. Anthropic
N.D. Cal. · Effective September 5, 2025 · Bartz v. Anthropic PBC, No. 3:24-cv-05417 (N.D. Cal.)
Authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson sued Anthropic over its use of pirated-book datasets to train Claude. In June 2025 Judge William Alsup ruled that training on lawfully purchased books was fair use. The piracy claims (LibGen ingestion) proceeded toward settlement; in September 2025 Anthropic agreed to a $1.5 billion class settlement covering ~500,000 works. Judge Alsup granted preliminary approval September 25, 2025 and final approval July 20, 2026. The appeal period runs through August 19, 2026; class members will not receive payments until at least end of 2026.
-
In effect
Thaler v. Perlmutter (Copyright)
D.C. Cir. · Effective March 18, 2025 · Thaler v. Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025)
The companion copyright case: Stephen Thaler sought to register a copyright with 'Creativity Machine' (his AI) as the author. The D.C. Circuit affirmed in March 2025 that the Copyright Act's human-authorship requirement is dispositive as a matter of statutory law. AI cannot be a copyright author under U.S. law.
-
Blocked / in litigation
Hachette v. Google (Gemini AI copyright)
United States · Effective July 10, 2026 · Hachette Book Group Inc. et al. v. Google LLC, No. 1:26-cv-05870 (S.D.N.Y., filed July 10, 2026)
Hachette Book Group, Cengage Learning, Elsevier, and author Scott Turow filed a copyright class action against Google in the Southern District of New York on July 10, 2026 (No. 1:26-cv-05870), alleging Google used millions of copyrighted books and academic journal articles without authorization to train its Gemini AI models. The complaint alleges Google misused works supplied for Google Books 'snippet' access to copy full texts for AI training, also sourcing material from pirate websites. An internal Google document cited in the complaint estimates potential copyright exposure of $10 billion to $100 billion. The case is distinct from (and does not overlap with) the existing NYT v. OpenAI and Authors Guild v. OpenAI suits.
-
Blocked / in litigation
Disney v. Midjourney
C.D. Cal. · Effective June 11, 2025 · Disney Enterprises, Inc. v. Midjourney, Inc., No. 2:25-cv-05275 (C.D. Cal.)
Disney and Universal — the first major Hollywood studios to sue a generative-AI company — filed a copyright action against Midjourney in June 2025 alleging Midjourney is a 'bottomless pit of plagiarism' that reproduces Star Wars, Marvel, Minions and other studio characters on demand.
-
Blocked / in litigation
CIR v. OpenAI
S.D.N.Y. · Effective June 27, 2024 · Ctr. for Investigative Reporting, Inc. v. OpenAI, Inc., No. 1:24-cv-04872 (S.D.N.Y.)
The nonprofit Center for Investigative Reporting (publisher of Mother Jones and Reveal) sued OpenAI and Microsoft in June 2024 over alleged use of its journalism for training. Joins the consolidated New York publisher actions.
-
Blocked / in litigation
UMG v. Suno
D. Mass. · Effective June 24, 2024 · UMG Recordings, Inc. v. Suno, Inc., No. 1:24-cv-11611 (D. Mass.)
The three major record labels — UMG, Sony Music, Warner — sued AI music generator Suno in Boston in June 2024, alleging Suno trained on copyrighted recordings and produces output that closely mimics specific tracks. A parallel suit was filed in S.D.N.Y. against competitor Udio.
-
Blocked / in litigation
UMG v. Udio
S.D.N.Y. · Effective June 24, 2024 · UMG Recordings, Inc. v. Uncharted Labs, Inc., No. 1:24-cv-04777 (S.D.N.Y.)
Companion to the Suno suit. UMG, Sony, and Warner sued Uncharted Labs (Udio) the same day, alleging Udio's training corpus included copyrighted master recordings and that outputs reproduce signature elements of identifiable tracks.
-
Blocked / in litigation
Daily News v. OpenAI
S.D.N.Y. · Effective April 30, 2024 · Daily News LP v. Microsoft Corp., No. 1:24-cv-03285 (S.D.N.Y.)
Eight Alden Global Capital newspapers (New York Daily News, Chicago Tribune, Orlando Sentinel, Sun Sentinel, San Jose Mercury News, Denver Post, Orange County Register, St. Paul Pioneer Press) sued OpenAI and Microsoft in April 2024 alleging mass copying of their journalism for GPT training and ChatGPT outputs.
-
Blocked / in litigation
NYT v. OpenAI / Microsoft
S.D.N.Y. · Effective December 27, 2023 · The New York Times Co. v. Microsoft Corp., No. 1:23-cv-11195 (S.D.N.Y.)
The New York Times sued OpenAI and Microsoft in December 2023, alleging the companies copied millions of Times articles to train GPT models and that ChatGPT regurgitates Times content verbatim. The case is the most consequential of the news-publisher AI training-data suits and is in discovery; in March 2025 Judge Sidney Stein largely denied OpenAI's motion to dismiss, allowing the direct, contributory, and DMCA claims to proceed. On July 9, 2026, the NYT and Daily News filed a motion to sanction OpenAI, alleging OpenAI concealed evidence. On August 6, 2026, Judge Stein denied the NYT's motion to amend its complaint to add Microsoft-specific contributory infringement claims, ruling the motion was untimely given the US Supreme Court's March 2026 ruling raising the bar for platform liability.
-
Blocked / in litigation
Concord Music v. Anthropic
M.D. Tenn. · Effective October 18, 2023 · Concord Music Group, Inc. v. Anthropic PBC, No. 3:23-cv-01092 (M.D. Tenn.); 5:24-cv-03811 (N.D. Cal.)
Concord, Universal Music Publishing, ABKCO, and other major music publishers sued Anthropic in Tennessee in October 2023, alleging Claude was trained on copyrighted song lyrics and reproduces them on demand. In March 2024 the court declined to grant a preliminary injunction; in 2025 the case was transferred to the Northern District of California where parallel actions are consolidated.
-
Blocked / in litigation
Authors Guild v. OpenAI
S.D.N.Y. · Effective September 19, 2023 · Authors Guild v. OpenAI Inc., No. 1:23-cv-08292 (S.D.N.Y.)
The Authors Guild plus 17 prominent fiction authors (George R.R. Martin, John Grisham, Jodi Picoult, Jonathan Franzen, Michael Connelly, and others) sued OpenAI alleging it copied their books wholesale to train GPT. Filed September 2023, the case was consolidated with related author suits under Judge Sidney Stein.
-
Blocked / in litigation
Silverman v. OpenAI
N.D. Cal. · Effective July 7, 2023 · Silverman v. OpenAI, Inc., No. 3:23-cv-03416 (N.D. Cal.)
Comedian and author Sarah Silverman, along with Christopher Golden and Richard Kadrey, sued OpenAI in July 2023 alleging ChatGPT was trained on their books via shadow-library datasets like Books3. The case proceeds in N.D. Cal. consolidated with related author actions.
-
Blocked / in litigation
Kadrey v. Meta
N.D. Cal. · Effective July 7, 2023 · Kadrey v. Meta Platforms, Inc., No. 3:23-cv-03417 (N.D. Cal.)
Authors Richard Kadrey, Sarah Silverman, Christopher Golden — later joined by Junot Díaz, Ta-Nehisi Coates, Andrew Sean Greer and others — sued Meta in July 2023 over the use of pirated books (LibGen / Books3) to train Llama. In June 2025 Judge Chhabria granted Meta partial summary judgment on fair use for these specific plaintiffs, but the ruling was narrow and the case continues on remaining theories.
-
Blocked / in litigation
Tremblay v. OpenAI
N.D. Cal. · Effective June 28, 2023 · Tremblay v. OpenAI, Inc., No. 3:23-cv-03223 (N.D. Cal.)
Authors Paul Tremblay and Mona Awad filed one of the first author copyright suits against OpenAI in the Northern District of California in June 2023, alleging ChatGPT was trained on their copyrighted books without permission. Now consolidated with Silverman, Chabon, and other West Coast author actions before Judge Araceli Martínez-Olguín.
-
Blocked / in litigation
Getty Images v. Stability AI (US)
D. Del. · Effective February 3, 2023 · Getty Images (US), Inc. v. Stability AI, Inc., No. 1:23-cv-00135 (D. Del.)
Getty Images sued Stability AI in Delaware in February 2023, alleging Stable Diffusion was trained on millions of Getty's copyrighted photos and that the model reproduces Getty's distinctive watermark. A separate parallel proceeding is underway in the UK High Court.
-
Blocked / in litigation
Andersen v. Stability AI
N.D. Cal. · Effective January 13, 2023 · Andersen v. Stability AI Ltd., No. 3:23-cv-00201 (N.D. Cal.)
Visual artists Sarah Andersen, Kelly McKernan, and Karla Ortiz brought the first major image-generator class action, alleging Stability AI, Midjourney, DeviantArt, and Runway trained text-to-image models on artists' copyrighted works scraped from the web. After two rounds of motions to dismiss, key claims survived and the case is in discovery.
-
Blocked / in litigation
Intercept v. OpenAI
S.D.N.Y. · Effective February 28, 2024 · Intercept Media, Inc. v. OpenAI, Inc., No. 1:24-cv-01515 (S.D.N.Y.)
The Intercept (and originally Raw Story and AlterNet, since voluntarily dismissed without prejudice) sued OpenAI on a DMCA-only theory — focusing on the alleged stripping of copyright-management information (author, title, terms) from training-data articles. The court allowed The Intercept's case to proceed in February 2025.
-
Blocked / in litigation
Chabon v. Meta
N.D. Cal. · Effective September 12, 2023 · Chabon v. Meta Platforms, Inc., No. 3:23-cv-04663 (N.D. Cal.)
The same Chabon plaintiff group filed a parallel action against Meta for Llama training, later consolidated into the broader Kadrey v. Meta proceedings.
-
Blocked / in litigation
Chabon v. OpenAI
N.D. Cal. · Effective September 8, 2023 · Chabon v. OpenAI, Inc., No. 3:23-cv-04625 (N.D. Cal.)
Pulitzer winner Michael Chabon, Ayelet Waldman, David Henry Hwang, and other authors sued OpenAI in September 2023 alleging ChatGPT was trained on their copyrighted books. Consolidated with Tremblay and Silverman before the same judge.
-
Blocked / in litigation
Doe v. GitHub (Copilot)
N.D. Cal. · Effective November 3, 2022 · Doe 1 v. GitHub, Inc., No. 4:22-cv-06823 (N.D. Cal.)
Anonymous software developers sued GitHub, Microsoft and OpenAI in November 2022, alleging GitHub Copilot trained on their open-source code in violation of open-source licences (which require attribution) and of DMCA § 1202. In September 2026 the U.S. Court of Appeals for the NINTH CIRCUIT REJECTED THE § 1202 THEORY, affirming the district court. Section 1202 prohibits intentionally removing copyright management information — an author's name, a copyright notice — from a copyrighted work. The developers argued that because code generated by the models resembled theirs but came out WITHOUT that information, the information had been unlawfully removed. The Ninth Circuit disagreed, and the holding is narrow but consequential: THE ABSENCE OF COPYRIGHT INFORMATION FROM A NEW WORK DOES NOT, BY ITSELF, MEAN SOMEONE ILLEGALLY REMOVED IT. Removing information from an existing work and creating a new work that never carried it are different acts. This is a LOSS for the plaintiffs and, on EFF's account, a win for programmers and internet users generally, because the opposite rule would have created a new source of liability reaching far past AI — remixes, classroom adaptations, reverse engineering — and would have fallen hardest on independent developers who cannot afford to litigate statutory-damages claims for years and would settle even when their use was lawful. Two things survive. Traditional copyright infringement claims are untouched: a developer can still sue if a model is used to reproduce their code. And the plaintiffs' CONTRACT claims against the AI companies, based on the open-source licence terms, remain live.
-
In effect
Copyright Office AI Guidance
United States · 17 U.S.C. § 102; U.S. Copyright Office, Copyright and AI Reports (2024–2025)
The U.S. Copyright Office has ruled that purely AI-generated works cannot be copyrighted — human creativity is required, and typing prompts alone is not enough. Its multi-part AI report covers digital replicas (2024), copyrightability of AI outputs (Jan 2025), and AI training on copyrighted works (May 2025 pre-publication). Whether AI training is fair use is being decided in ongoing litigation.
-
Blocked / in litigation
Thomson Reuters v. Ross
D. Del. · Effective February 11, 2025 · Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence, Inc., No. 1:20-cv-00613 (D. Del. Feb. 11, 2025)
Thomson Reuters sued legal-research startup Ross Intelligence in 2020 for copying Westlaw headnotes to train a competing AI legal-research tool. In February 2025, Judge Stephanos Bibas (sitting by designation) granted summary judgment to Thomson Reuters on direct copyright infringement and rejected Ross's fair-use defense — the first definitive U.S. ruling on AI-training fair use. No jury trial occurred: a 2023 opinion had denied summary judgment and pointed toward trial, but the court invited renewed briefing and reversed course in the 2025 ruling.
-
Repealed / replaced
USPTO AI Inventorship Guidance
United States · Effective February 13, 2024 · 89 Fed. Reg. 10043 (Feb. 13, 2024) — rescinded Nov. 28, 2025
USPTO guidance confirming AI-assisted inventions are patentable but inventorship analysis focuses on significant human contributions (Pannu factors). Rescinded November 28, 2025 and replaced by 2025 successor guidance.
State copyright and training data rules (6)
-
In effect
ELVIS Act
Tennessee · Effective July 1, 2024 · Tenn. Code Ann. §§ 47-25-1101 to -1108 (ELVIS Act, 2024)
The first US law protecting voices from AI cloning: Tennessee added 'voice' to its right-of-publicity law, so using AI to mimic someone's voice or likeness without permission is both a civil violation and a crime. It also allows lawsuits against those who distribute tools whose primary purpose is producing unauthorized voice or likeness replicas.
-
In effect
Plateau Peoples TK/BC Labels
Plateau Peoples' Web Portal (Multi-Tribal) · Effective January 1, 2015 · Plateau Peoples' Web Portal — multi-tribal TK/BC Labels initiative
Six Plateau tribes — Colville, Umatilla, Warm Springs, Yakama, Spokane, and Coeur d'Alene — jointly implement Local Contexts Traditional Knowledge and Biocultural Labels on digital cultural-heritage collections. A working Indigenous data sovereignty mechanism applicable to AI training data: labels travel with the data and assert community-defined access and use rules.
-
In effect
SSB 5886 (WA Digital Likeness Rights)
Washington · Effective June 11, 2026 · Wash. SSB 5886, Ch. 69, 2026 Laws; RCW ch. 63.60 (amending personality rights statute)
Washington updated its personality-rights law so that AI-generated audio or video that realistically mimics someone's face or voice without consent — called a 'forged digital likeness' — is now a civil violation. Victims can seek court injunctions to stop the misuse, and the civil penalty for each infringement is $3,000 plus any actual damages they can prove. The law became effective June 11, 2026.
-
In effect
AB 2013 (Training Data Transparency)
California · Effective January 1, 2026 · Cal. Civ. Code §§ 3110–3111 (AB 2013, Stats. 2024)
Developers of generative AI systems made available to Californians must publicly post documentation about the datasets used to train their models, including sources, whether they contain personal information or copyrighted material, and time periods of collection. Applies to systems released or substantially modified since January 1, 2022.
-
In effect
AR Generative AI Ownership Act (Act 927)
Arkansas · Effective August 5, 2025 · Ark. Act 927 (2025) (HB1876), codified at Ark. Code Ann. tit. 18, ch. 4 (Sec. 18-4-101 et seq.)
Arkansas set default ownership rules for the inputs and outputs of generative AI tools. A person who supplies the input or direction to a generative AI tool owns the content it produces, as long as that content does not infringe existing intellectual property rights. A person who lawfully supplies the data used to train a model owns the resulting trained model. When an employee is directed to use a generative AI tool within the scope of their job, the employer owns the resulting output and model. The law sets property rights rather than prohibitions, so it does not impose penalties.
-
Proposed / pending
New York Stealth Crawler Prohibition Act (2026)
New York · N.Y. S. 9934 / A. 11292 (2025-2026 Sess.) — New York Stealth Crawler Prohibition Act (passed legislature; awaiting gubernatorial action)
New York's Legislature has passed the Stealth Crawler Prohibition Act, S9934 / A11292, and sent it to Governor Kathy Hochul. It is the first bill of its kind in the country. The rule is simple to state: any web crawler that accesses a covered news source must disclose its identity and its purpose, either before or at the time of access. Deploying a 'stealth crawler' in a way that damages or economically harms a news publisher is a civil violation carrying penalties of up to $15,000 PER DAY. The bill is framed as protection for local journalism against AI companies scraping news content without identifying themselves. It has drawn organized opposition from an unusual direction. The Electronic Frontier Foundation, the Center for Democracy & Technology and more than a dozen other civil society organizations — nineteen groups in all — wrote to Governor Hochul urging a veto. Their argument is that although the bill is presented as a journalism measure, it effectively DEANONYMIZES AND CRIMINALIZES AUTOMATED ACCESS TO THE OPEN WEB: it would threaten anonymous access to information, compel speculative disclosures about purpose, and expose researchers, journalists, watchdogs and ordinary internet users to liability simply for using automated tools to read publicly available pages. This entry is recorded as PROPOSED. The bill is not law, the Governor had not acted as of mid-September 2026, and action was widely expected to wait until after the November 7, 2026 gubernatorial election.