Today a site that wants to stay out of AI training has to list each company's crawler: GPTBot, Google-Extended and the rest. The draft (draft-kelley-aipref-training-preference-signal-00, written by Kevin Kelley, a standards lead at Anthropic, according to The Verge) adds one shared name instead. You write a normal robots.txt group with User-Agent: __usage_ai-training and use Disallow and Allow paths as usual. The draft adds no new directives, and parsers that don't know the token skip the group like any unknown user agent.
OpenAI said on October 30 it will honor the signal and aims to implement it in the coming months, alongside its existing crawler controls. Google said that when a site opts out with the token and doesn't name Google-Extended, it will treat that as a Google-Extended disallow, and it aims to support it in 2027. Google-Extended still works on its own, and sites that allow it can still show up through grounding in Gemini Apps. The Verge reports that Microsoft also signed on (with separate controls for training and grounding, and no effect on Bing results), that Anthropic already honors it, and that Meta voiced support without saying it would comply.