A new Google white paper stakes out a narrow theory of copyright liability for generative AI, arguing that ingestion and training don't infringe—only the outputs can. That framing matters enormously for data licensing markets: if courts adopt it, rightsholders lose leverage over how their works are scraped and used to build models, and pressure shifts entirely to output-side enforcement.
Expect plaintiffs in pending AI training-data suits to push back hard on this input/output distinction as it works its way into litigation briefing.
Google White Paper Says Only AI Outputs Infringe Copyright
