Unsealed Court Docs Reveal AI Execs Warned of 'Theft' and 'Doom Loop' in News Training

Key points

• Microsoft Director of Applied Science Brent Hecht called AI training an 'astonishing theft' and a 'complete mockery' of fair use in internal documents.
• OpenAI’s ChatGPT head Nick Turley warned that AI products are 'largely substitutive' for news, creating an 'existential threat' to publishers.
• Microsoft data showed 83-93% lower click-through rates for news sites when users used AI answer tools compared to traditional search.
• OpenAI President Greg Brockman reportedly responded 'ah nice' when informed of a hack to bypass the New York Times paywall for training data.
• News plaintiffs argue these internal admissions undermine the defendants' fair use defense, as AI outputs replace the original news content.

Background

This development follows earlier coverage of the New York Times’ lawsuit against Microsoft and OpenAI, which centers on whether training AI models on copyrighted news content constitutes fair use. Previous reports highlighted the tension between AI companies’ reliance on public data and publishers’ claims of unauthorized copying. The unsealing of these documents marks a significant escalation, exposing internal dissent and strategic concerns within the tech giants regarding the long-term sustainability of their AI training methods.

How outlets are covering it

The Washington Post and Ars Technica emphasize the stark contrast between the companies' public fair use defense and their private internal warnings about 'theft' and a 'doom loop' that could harm AI performance by destroying news revenue. The Financial Times highlights the financial motivations, noting OpenAI co-founder Greg Brockman’s focus on 'gazillions' in revenue despite knowing the 'existential threat' to publishers. EFF, while not directly covering this specific lawsuit, provides a broader context on copyright law, arguing that courts should not expand copyright liability beyond what Congress authorized, as seen in their recent victory against a DMCA claim involving AI training on GitHub code. This suggests a legal landscape where AI companies face both direct copyright infringement claims and broader challenges to the scope of digital copyright protections.

Why it matters

The unsealed documents provide critical evidence that AI companies were aware their training methods could undermine the news industry that supplies their data. This could significantly impact the fair use defense in ongoing and future copyright lawsuits, potentially forcing AI companies to license content or face substantial damages. It also raises questions about the long-term sustainability of AI models that rely on content from industries they may be disrupting.

What to watch

The New York Times and other news plaintiffs are preparing for trial, aiming to use these internal admissions to prove that AI outputs substitute for news content, thereby defeating the fair use defense. Microsoft and OpenAI are expected to continue defending their actions as transformative fair use, but the unsealed documents may complicate their legal strategy. The outcome of this case could set a precedent for how AI companies interact with copyrighted content in the future.