Newly unsealed court documents from the ongoing copyright lawsuit brought by The New York Times against OpenAI and Microsoft reveal internal admissions that the companies’ AI training methods amounted to theft. According to the filings, a top Microsoft executive described the mass scraping of paywalled content as “the largest theft of labor in human history,” highlighting the vast scale of unauthorized copying involved. Both companies reportedly bypassed paywalls without detection and intentionally removed copyright notices from the training data, which included hundreds of thousands of articles from The Times and other news publishers.

The documents also disclose that Microsoft’s AI-powered Bing Copilot caused drastic declines in traffic to paywalled news sites, with click-through rates dropping by as much as 93%, a situation Microsoft termed a damaging “doom loop” that threatens the sustainability of content creators. Microsoft CEO Satya Nadella testified that all paywalled content should be licensed for AI training, and asserted he would have enforced re-training of models had he known OpenAI scraped protected material. OpenAI’s own leadership privately acknowledged these AI models pose an existential threat to publishers and journalists by substituting traditional news consumption.

According to the filings, OpenAI’s datasets incorporated over 91,000 copies of works from The Times and other news organizations, obtained via the Bing Index and initiatives like Project Taxi and Project Mango, which exchanged content back and forth between the tech giants. An internal OpenAI conversation revealed excitement over techniques to circumvent The New York Times paywall. The companies’ actions also breached fair use principles by using material in ways that compete directly with the original content, rather than transforming it in a non-substitutive manner.

The revelations deepen scrutiny over how AI firms train language models using copyrighted content without licensing, a legal gray area with courts frequently siding with AI companies’ fair use arguments. However, the newly exposed internal communications illustrate a recognition within Microsoft and OpenAI that their practices could undermine the business models of news publishers and possibly violate copyright protections. Both organizations have declined to comment on the filings as the case continues to unfold.