Third Circuit rules AI training on copyrighted material isn’t fair use
copyright training
| Source: Mastodon | Original article
The U.S. Third Circuit ruled that using copyrighted text to train AI models does not qualify as fair use, marking a major shift in copyright law for LLM training.
The U.S. Court of Appeals for the Third Circuit has ruled that training an artificial‑intelligence system on copyrighted material does not qualify as fair use. In a unanimous decision issued on 29 September 2026, the panel affirmed a partial summary‑judgment in Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., holding that the AI startup’s copying of 2,243 Westlaw headnotes to train its legal‑research platform infringed Thomson Reuters’ copyrights.
The court found the headnotes “original enough for copyright protection” and concluded that three of the four statutory fair‑use factors weighed against ROSS Intelligence. The decision marks the first federal appellate ruling directly addressing whether the mass extraction of copyrighted text for AI training can be shielded by the fair‑use doctrine.
The judgment is a watershed for the AI industry. By rejecting a core defense that many developers have relied on to justify the use of large text corpora, the ruling could force companies to renegotiate data licences, redesign training pipelines, or limit the scope of models that draw on proprietary content. Legal‑tech firms, in particular, may need to reassess how they build and market products that depend on curated legal databases.
The case is likely to prompt further litigation as AI developers test the boundaries of the decision. Observers will watch whether Thomson Reuters seeks review by the Supreme Court, how other jurisdictions respond, and whether lawmakers introduce clarifying statutes. The outcome will shape the balance between copyright protection and the rapid expansion of generative AI across sectors.
Sources
Back to AIPULSEN