Microsoft says virtually nobody was grabbing NYT articles through its chatbot

Editor's take

Microsoft claims its Copilot AI model, specifically the underlying LLM, demonstrably avoids verbatim reproduction of copyrighted material from sources like The New York Times. This assertion, made within legal filings, directly challenges publisher lawsuits alleging widespread copyright infringement by AI models, such as those brought by the NYT, which accuse models like OpenAI's GPT-4 of generating derivative works.

The significance lies in the potential for this technical defense to reshape the legal landscape surrounding AI training data and output. If Microsoft can effectively prove that Copilot's outputs are sufficiently transformative and do not constitute direct infringement, it could provide a blueprint for other AI developers facing similar litigation, potentially de-escalating the current publisher-versus-AI showdown.

Future developments to monitor include independent audits of Copilot's output and the legal system's interpretation of "fair use" in the context of generative AI. The precise threshold for what constitutes "substantive chunks" versus acceptable paraphrasing will be critical, and a ruling that favors Microsoft's technical defense could embolden broader adoption of AI models trained on vast, potentially copyrighted, datasets.