Sealed NYT briefs show OpenAI, Microsoft staff privately doubted AI scraping

Legal filings in The New York Times’ copyright suit against OpenAI and Microsoft, surfaced by 404 Media from documents both companies wanted sealed, reveal that leaders at both firms internally described AI training practices in stark terms. Microsoft Applied Science director Brent Hecht wrote in a January 2023 memo that large models ‘hoovering up’ people’s work amounted to ‘the largest theft of labor in human history,’ noting that almost no one who created the content consented to this use or was paid for it. He also warned of a ‘doom loop’ in which AI products cannibalize the very publishers they depend on, calling it unusual for an end product to threaten ‘the economic foundations of its essential suppliers.’

The brief bolsters that concern with data: Microsoft’s own numbers reportedly showed Copilot cut click-through rates to the Times by as much as 93% versus Bing search. On the OpenAI side, ChatGPT head Nick Turley called the chatbot an ‘existential threat’ to publishers because it is ‘largely substitutive,’ an engineer testified that users won’t click source links no matter how prominently they’re shown, and researcher Nick Ryder floated a ‘hack to get around’ the Times paywall — to which president Greg Brockman replied, ‘ah nice.’

The disclosures cut against the fair-use defense both companies are leaning on, since fair use weighs the effect of copying on a work’s market value, and these communications show executives were aware of that harm. Even Microsoft CEO Satya Nadella conceded in a deposition that paywalled content should be licensed for training, and said he would have demanded OpenAI retrain its models had he known it scraped material behind paywalls. The stakes are amplified by a separate ruling that found Anthropic’s use of published material qualified as fair use — leaving the doctrine’s application to AI training unsettled as the Times pushes for summary judgment.