Market & Regulation
OpenAI and Microsoft internal docs admit web 'doom loop'
The Verge AIhigh confidence
Why it matters
- →Internal documents confirm AI companies anticipated the collapse of publisher revenue as a direct consequence of their training strategies.
- →Executives acknowledged that replacing search traffic creates a feedback loop that degrades the quality of future model training data.
- →The filings provide concrete evidence that copyright infringement and memorization were known technical risks, not just legal liabilities.
Newly unsealed documents from the New York Times’ lawsuit against OpenAI and Microsoft expose internal warnings about the long-term damage AI training causes to the web. Executives including Satya Nadella and Brent Hecht acknowledged that their models were creating a "doom loop" by replacing search traffic, thereby destroying the economic incentives for content creation. The filings also detail concerns over copyright infringement, with employees noting that systems like GPT-4 memorize and regurgitate copyrighted material verbatim. These admissions suggest that both companies proceeded with large-scale web scraping despite knowing it threatened their own supply chain of high-quality data.
Read originalThe story around this
TopicAnthropic Pentagon Legal DisputeCooling
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
