Skip to content

Generated Source Mirror

This page is generated from the source Markdown file so it can be viewed inside the MkDocs site.

📝 Page Status: Draft — sourced but not yet reviewed

  • Status: draft
  • Sources: 15
  • Relationships: 17
  • Research debt items: 5

Copyright and Generative AI

Summary

U.S. copyright law is being applied to generative AI along two tracks. Outputs: courts and the Copyright Office currently treat purely machine-determined expression as uncopyrightable. Inputs: district courts have split on whether copying works to train models is fair use, and they have distinguished pirated acquisition, competing non-generative tools, and news-output substitution.

Verified Facts

  • Copyright protection under U.S. law for purposes such as criticism, comment, news reporting, teaching, scholarship, or research is limited by the four fair-use factors in 17 U.S.C. § 107.1
  • The Copyright Office’s March 16, 2023 registration guidance states that copyright can protect only material that is the product of human creativity, that applicants must disclose more-than-de-minimis AI-generated content, and that when AI determines the expressive elements of an output, that material is not protected by copyright.2
  • The same guidance states that humans may claim copyright in their own contributions, including creative selection, coordination, and arrangement of AI-generated material or sufficiently creative modifications, and that applicants should not list an AI system as author merely because they used it.2
  • The Office’s January 2025 copyrightability report concluded that existing law can resolve AI copyrightability questions without legislative change; that purely AI-generated material is not copyrightable; that prompts alone do not provide sufficient control on current generally available technology; and that the case has not been made for sui generis protection of AI-generated content.3
  • In Thaler v. Perlmutter, the D.C. Circuit held that the Copyright Act requires eligible works to be authored in the first instance by a human being and affirmed refusal to register an image listed as authored solely by a generative system.4
  • In Thomson Reuters v. Ross Intelligence, the District of Delaware rejected fair use for a competing non-generative legal-research AI trained on Westlaw headnotes.5
  • In Bartz v. Anthropic PBC, the Northern District of California held that training Claude on books was transformative fair use but that a permanent library of pirated copies was not excused as fair use; a $1.5 billion class settlement of the piracy claims received final approval on July 20, 2026.67
  • In Kadrey v. Meta Platforms, Inc., the Northern District of California granted Meta summary judgment on fair use for training Llama on thirteen authors’ books on that record, while warning that market-dilution evidence could change the result in other cases and leaving torrenting-distribution claims open.8
  • In The New York Times Company v. Microsoft Corp., the Southern District of New York in April 2025 largely denied motions to dismiss publisher copyright claims against OpenAI and Microsoft; an August 2026 order later dismissed material-contribution contributory claims premised on alleged end-user infringement.9
  • In Spyder Games LLC v. Mementum Lab, game studios seek a declaration that AI-generated “brainrot” meme characters are not copyrightable. See Brain Rot and AI-Generated Meme Characters.10
  • Music-industry suits are collected in Music Copyright and Generative AI: labels allege unlicensed copying of sound recordings to train Suno and Udio; publishers allege lyric copying in Concord v. Anthropic; and GEMA v. OpenAI is a non-final German first-instance judgment that memorized lyrics in models 4 and 4o were unauthorized reproductions.11121314

Historical Context

Software copyright already had a fair-use landmark in Google LLC v. Oracle America, Inc., which treated copying of Java API declaring code as transformative fair use. Generative-AI cases cite Google v. Oracle but present a different use: ingesting expressive works at scale to produce competing text, images, or music. No U.S. court of appeals had, as of the sources reviewed here, issued a merits holding that LLM training on copyrighted books is or is not fair use.

Three questions should be kept separate.

Authorship of outputs. Thaler and Copyright Office practice answer the easy case: a work with no human author cannot be registered or owned as copyrighted expression.234 Mixed works remain a case-by-case inquiry. Spyder tests whether viral meme characters created with image generators fall on the uncopyrightable side of that line.10

Copying for training (inputs). Fair use under § 107 is fact-specific.1 Ross involved a substitute legal-research tool, not a generative writer.5 Bartz treated generative training as transformative but piracy-for-library as infringement.6 Kadrey treated generative training as fair use on a thin market-harm record while saying other records may come out the other way.8 Publisher cases such as the Times action add output-substitution and secondary-liability theories that book-training summary judgments did not decide.9 Music cases add a further split: sound-recording training claims against Suno and Udio remain without a U.S. fair-use holding, while GEMA is a non-final German holding that lyric memorization is reproduction outside the TDM exception.111214

Remedies and settlements. Bartz’s $1.5 billion settlement prices past pirated acquisition for a defined class; it does not freeze the fair-use doctrine or release output claims.7

Non-copyright AI suits (chatbot product liability, defamation, privacy) are collected in AI Litigation Landscape.

Relationships

  • TOPIC-AI-COPYRIGHT cites SRC-USCO-AI-COPYRIGHTABILITY-REPORT-2025.
  • TOPIC-AI-COPYRIGHT related_to CASE-THALER-V-PERLMUTTER.
  • TOPIC-AI-COPYRIGHT related_to CASE-BARTZ-V-ANTHROPIC.
  • TOPIC-AI-COPYRIGHT related_to CASE-KADREY-V-META.
  • TOPIC-AI-COPYRIGHT related_to CASE-THOMSON-REUTERS-V-ROSS.
  • TOPIC-AI-COPYRIGHT related_to CASE-NYT-V-OPENAI.
  • TOPIC-AI-COPYRIGHT related_to CASE-SPYDER-V-MEMENTUM.
  • TOPIC-AI-COPYRIGHT related_to TOPIC-AI-MUSIC-COPYRIGHT.
  • TOPIC-AI-COPYRIGHT related_to CASE-UMG-V-SUNO.
  • TOPIC-AI-COPYRIGHT related_to CASE-UMG-V-UDIO.
  • TOPIC-AI-COPYRIGHT related_to CASE-CONCORD-V-ANTHROPIC.
  • TOPIC-AI-COPYRIGHT related_to CASE-GEMA-V-OPENAI.
  • TOPIC-AI-COPYRIGHT related_to STAT-USC-107.
  • TOPIC-AI-COPYRIGHT related_to TOPIC-BRAIN-ROT.
  • TOPIC-AI-COPYRIGHT related_to TOPIC-AI-LITIGATION.
  • TOPIC-AI-COPYRIGHT related_to CASE-GOOGLE-V-ORACLE.
  • TOPIC-AI-COPYRIGHT related_to PERSON-KIT-WALSH.

Sources

  1. SRC-USC-17-107-LII: 17 U.S.C. § 107.
  2. SRC-USCO-AI-REGISTRATION-GUIDANCE-2023: Copyright Office AI registration guidance (2023).
  3. SRC-USCO-AI-COPYRIGHTABILITY-REPORT-2025: Copyright Office AI copyrightability report, Part 2 (2025).
  4. SRC-THALER-V-PERLMUTTER-CADC: Thaler v. Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025).
  5. SRC-THOMSON-REUTERS-V-ROSS-2025: Thomson Reuters v. Ross Intelligence (D. Del. 2025).
  6. SRC-BARTZ-V-ANTHROPIC-FAIR-USE: Bartz v. Anthropic, Order on Fair Use (2025).
  7. SRC-BARTZ-V-ANTHROPIC-FINAL-APPROVAL: Bartz settlement final approval (2026).
  8. SRC-KADREY-V-META-FAIR-USE: Kadrey v. Meta (N.D. Cal. 2025).
  9. SRC-NYT-V-OPENAI-MTD-2025: Times v. Microsoft/OpenAI MTD opinion (2025).
  10. SRC-SPYDER-V-MEMENTUM-COMPLAINT: Spyder Games v. Mementum complaint (2025).
  11. SRC-UMG-V-SUNO-COMPLAINT: UMG v. Suno complaint (2024).
  12. SRC-UMG-V-UDIO-COMPLAINT: UMG v. Udio complaint (2024).
  13. SRC-CONCORD-V-ANTHROPIC-MTD-2025: Concord v. Anthropic MTD denial (2025).
  14. SRC-GEMA-V-OPENAI-LG-MUENCHEN-2025: GEMA v. OpenAI, LG München I (2025).

Additional sources (not yet cited in footnotes):

Research Debt

  • Add Getty Images v. Stability AI (UK High Court 2025 and N.D. Cal. U.S. action) with primary judgments.
  • Add Andersen v. Stability AI / artists’ class actions from complaints and orders.
  • Add Copyright Office Part 3 (training) when a final version is cited from copyright.gov rather than secondary summaries.
  • Document U.S. Copyright Office registration examples beyond Thaler (Zarya of the Dawn; later granted mixed-work registrations).
  • Music-generator and lyric cases now have dedicated pages; add later U.S. fair-use orders and any OLG München appeal in GEMA there.

Document metadata

  • Last verified: 2026-08-22

  1. SRC-USC-17-107-LII — 17 U.S.C. § 107. Primary statutory authority. ↩↩

  2. SRC-USCO-AI-REGISTRATION-GUIDANCE-2023 — 88 Fed. Reg. 16190. Agency guidance. ↩↩↩

  3. SRC-USCO-AI-COPYRIGHTABILITY-REPORT-2025 — Copyright Office Part 2 report (Jan. 2025). Agency report. ↩↩

  4. SRC-THALER-V-PERLMUTTER-CADC — 130 F.4th 1039 (D.C. Cir. 2025). Primary appellate authority. ↩↩

  5. SRC-THOMSON-REUTERS-V-ROSS-2025 — D. Del. Feb. 11, 2025. District-court authority. ↩↩

  6. SRC-BARTZ-V-ANTHROPIC-FAIR-USE — N.D. Cal. June 23, 2025. District-court authority. ↩↩

  7. SRC-BARTZ-V-ANTHROPIC-FINAL-APPROVAL — N.D. Cal. July 20, 2026. Settlement approval order. ↩↩

  8. SRC-KADREY-V-META-FAIR-USE — 788 F. Supp. 3d 1026 (N.D. Cal. 2025). District-court authority. ↩↩

  9. SRC-NYT-V-OPENAI-MTD-2025 — S.D.N.Y. Apr. 4, 2025. Pleading-stage authority. ↩↩

  10. SRC-SPYDER-V-MEMENTUM-COMPLAINT — N.D. Cal. Nov. 26, 2025. Pending case; plaintiff allegations. ↩↩

  11. SRC-UMG-V-SUNO-COMPLAINT — D. Mass. June 24, 2024. Pending; plaintiff allegations. ↩↩

  12. SRC-UMG-V-UDIO-COMPLAINT — S.D.N.Y. June 2024. Pending; plaintiff allegations. ↩↩

  13. SRC-CONCORD-V-ANTHROPIC-MTD-2025 — N.D. Cal. Oct. 6, 2025. Pleading-stage order. ↩

  14. SRC-GEMA-V-OPENAI-LG-MUENCHEN-2025 — LG München I Nov. 11, 2025. Non-final first-instance judgment. ↩↩