Court Files: OpenAI Knew Its AI Was Draining the Web Dry

Newly unsealed court documents from the New York Times' lawsuit against OpenAI and Microsoft reveal something striking: the companies knew exactly what they were doing to the web, and did it anyway. In internal memos, Microsoft's Brent Hecht called AI's use of news articles and other writing 'the largest theft of labor in human history,' and said the company's legal defense makes a 'complete mockery' of fair use — the rule that lets people quote and reuse small bits of copyrighted work.

Maybe the most eye-opening quote comes from Microsoft's own internal planning: their AI content strategy had 'started a doom loop that will hurt the performance of our models and the entire web at the same time.' In plain terms, a doom loop means AI answers your question so you never visit the website that wrote the article — so that site earns less money, writes fewer articles, and the AI has less good material to learn from. It noted it's 'highly unusual that an end-product threatens the economic foundations of its essential suppliers.' That's like a restaurant chain putting all its farmers out of business, then wondering where the food went.

The documents also show OpenAI staff admitting that GPT-4 'memorized a ton of data' and was 'insanely good at regurgitation' — AI-speak for copying chunks of text word-for-word. In testing, ChatGPT spat out long passages straight from the Times, Mercury News, The Denver Post, LifeHacker, and Eurogamer. One Microsoft memo admitted that 'almost no one intended' their content to be used this way, 'nor are they compensated for its use.'

Microsoft now says these were one employee's personal views, not company policy. But the outcome is hard to argue with: searches increasingly end in an AI answer, not a click. Whether you write, read, or run a small business online, this fight decides who gets paid when machines learn from human work.