A US federal appellate court upheld on 29 September a lower-court ruling that Ross Intelligence infringed copyright by training its legal AI system on Thomson Reuters' Westlaw headnotes without authorisation. The ruling establishes that using copyright-protected material to train an AI model is not automatically shielded by fair use — the most consequential judicial precedent in AI training data law to date, directly applicable to every AI company that has trained on crawled publisher content. In a separate development the same day, US newspapers suing OpenAI publicly criticised the Justice Department's stated position that LLM training on copyrighted content qualifies as fair use, describing it as a stance that "favours AI at the expense of publishers." The appellate ruling gives that litigation significantly stronger precedent to draw on.
Google's AI Content Pilot Is Paying Most Publishers Under 0.1% of Their Ad Revenue
The Information, 29 September: Google's ~100-publisher pilot for content used in Gemini, AI Overview…