Policy
Courts Split on AI Training Copyright, Leaving Congress to Decide
Conflicting federal rulings on fair use have created legal uncertainty for AI companies using copyrighted material to train models.
Key takeaways
- Federal district courts in California ruled AI training on copyrighted works was fair use, while the Third Circuit reached the opposite conclusion in a legal research case.
- More than 150 copyright lawsuits have been filed against AI companies in the United States amid legal uncertainty.
- The Trump administration opposes congressional action and recommends letting courts resolve whether AI training constitutes fair use.
- Five bills addressing AI training and copyright remain stalled in committee with no advancement beyond referral.
- Courts and the Copyright Office require human authorship for copyright protection, rejecting claims for works generated entirely by AI systems.
Competing court rulings create uncertainty
Federal courts have issued contradictory decisions on whether artificial intelligence companies can legally train their models on copyrighted material without permission, according to a Congressional Research Service report. The legal uncertainty has generated more than 150 copyright lawsuits against AI companies in the United States and left Congress facing a choice: establish new statutory protections for copyright holders or allow the judiciary to continue developing precedent case by case.
Two June 2025 district court decisions in California's Northern District—Bartz v. Anthropic and Kadrey v. Meta Platforms—both found that the AI training practices at issue qualified as fair use. However, the judges differed in their analysis of market harm, examining whether AI-generated content could compete with the original copyrighted works.
The Bartz decision also distinguished between training on lawfully obtained books versus pirated copies. Anthropic subsequently agreed to pay $1.5 billion to settle claims involving its use of pirated materials, with final approval granted in July.
Third Circuit establishes appellate precedent
A September 2025 ruling from the U.S. Court of Appeals for the Third Circuit reached the opposite conclusion. In a case involving ROSS Intelligence, which copied Westlaw case summaries to train a legal research system, the court ruled the copying was not fair use. The panel found ROSS used the material to build a competing commercial product, that the copying was minimally transformative, and that it harmed Westlaw's potential licensing market.
The Third Circuit expressly distinguished its decision from generative AI cases, noting that ROSS developed a legal research platform rather than a system that generates original content. This distinction may limit the ruling's application to companies building large language models or image generators.
Administration favors judicial resolution
The Trump administration has taken a clear position against congressional intervention. The White House's National Policy Framework for Artificial Intelligence, released in March, states the administration believes training AI models on copyrighted material does not violate copyright law, though it acknowledges competing legal arguments. The framework recommends Congress avoid legislation that would interfere with courts resolving the fair-use question.
The Department of Justice reinforced this stance in September by filing a brief supporting OpenAI's position in a copyright dispute.
Five bills pending in committee
Congress is considering multiple legislative approaches. Five bills introduced in the 119th Congress remain in committee:
- The COPIED Media Act (S. 1396) would establish civil liability for certain uses of copyrighted material in AI training
- The AI Accountability and Personal Data Protection Act (S. 2367) would create additional liability provisions
- The TRAIN Act (S. 2455 and H.R. 7209) would allow copyright owners to subpoena AI developers to disclose whether their works were used in training
- The CLEAR Act (S. 3813) would require disclosures to the Copyright Office about copyrighted works used in training
None has advanced beyond committee referral.
Why it matters
The legal uncertainty affects billions of dollars in AI investment and the viability of business models built on training large models with internet-scale datasets. Without clear rules, AI companies face unpredictable litigation risk, while copyright holders lack consistent remedies when their works are used without permission. The outcome will shape whether the United States maintains permissive AI development policies or adopts stricter protections similar to those being considered in Europe.
Copyright Office maintains human authorship requirement
On the separate question of whether AI-generated material can receive copyright protection, courts and the Copyright Office have established that human authorship is required. The D.C. Circuit ruled in Thaler v. Perlmutter that works attributed entirely to AI systems cannot be copyrighted, a decision the Supreme Court declined to review in March. The Copyright Office's 2023 guidance maintains that AI-generated material is not independently eligible for protection, though works containing AI elements may qualify when they include sufficient original human authorship.
These details were first reported by Legis1, which published a comprehensive analysis based on the Congressional Research Service report.