Google, Microsoft, OpenAI and Anthropic support a unified AI training opt-out mechanism
Google, Microsoft, OpenAI and Anthropic back one AI training opt-out
On October 30, 2026, Google, Microsoft, OpenAI and Anthropic announced they will support a unified robots.txt policy marker, __usage_ai-training, so that a website can declare once and simultaneously refuse all participating vendors from using its content for generative AI training. The marker originates from the IETF individual draft draft-kelley-aipref-training-preference-signal-00, submitted the same day by Anthropic engineer Kevin Kelley; the draft has no formal status in the IETF and expires on May 3, 2027. Microsoft is set to enforce it starting February 1, 2027, Google targets early 2027, OpenAI says it will implement it within months, and Anthropic has begun development but has not given a date. The wildcard User-agent: * does not constitute a training preference declaration, and the marker only constrains content use, not crawling behavior.
The same training opt-out flag also withdraws Gemini answer citations on Google, whereas on Microsoft it is controlled separately from answer citations; website operators need to check each company's settings separately.