AI companies, such as OpenAI, Google, and Meta, assert that their large language models (LLMs) do not store copyrighted content but rather learn from their training data akin to human cognition. This distinction is pivotal in their defense against increasing legal challenges related to copyright law, which protects original creative works under the U.S. Copyright Act of 1976. The industry’s reliance on the “fair use” doctrine for criticism and research has faced scrutiny from rights holders who argue that companies exploit copyrighted materials without fair compensation. A recent study by Stanford and Yale researchers reveals that certain LLMs, including GPT-4.1 and Google’s Gemini 2.5 Pro, reproduce copyrighted texts nearly verbatim, challenging the companies’ claims. With legal battles intensifying, these findings could significantly impact the AI industry’s defense strategies and financial liabilities, highlighting the urgent need for robust discussions on copyright implications in AI development.
Source link
