Unsealed documents show OpenAI and Microsoft were aware of potential "doom loop" for the web

Internal documents from Microsoft describe an AI content strategy that has initiated a "doom loop," potentially damaging both the performance of AI models and the entire web. Microsoft noted that it is "highly unusual" for an end-product to threaten the economic foundations of its essential suppliers, a situation they identified regarding their LLM (Large Language Model) business and the content supply chain.

Microsoft's Director of Applied Science, Brent Hecht, characterized the scraping of data to train models as "the largest theft of labor in human history" and stated that such practices make a "complete mockery of the idea of fair use." While Microsoft spokespeople have attempted to distance the company from these specific individual comments, the filings include perspectives from various figures within both companies.

Internal OpenAI documents also indicated awareness of the tendency of models like GPT-4 to reproduce copyrighted material verbatim, with employees noting the model's high capacity for "regurgitation." Additionally, OpenAI's own experts attributed the decline in referral traffic for news sites to AI summaries, speculating that search referrals could drop by as much as 60 percent.

PLUS ULTRAby Amenoyomi

The "doom loop" refers to a structural risk where AI models undermine the economic foundations of the content creators who serve as their essential suppliers. Internal Microsoft documentation explains that when AI chatbots provide direct answers, they replace the need for users to visit the original source, effectively cutting off the referral traffic and revenue that sustain the web's content supply chain.