Four AI giants back one opt-out signal: a single robots.txt rule to refuse AI training across OpenAI, Google, Microsoft and Anthropic

SAN FRANCISCO — OpenAI, Google, Microsoft and Anthropic announced on October 30 that they will support a single shared signal for opting out of AI training. A website operator that adds the policy token __usage_ai-training to its robots.txt file will be able to tell all four companies' crawlers at once that it does not want its content used to train AI models, instead of maintaining separate rules for each crawler.

One rule instead of four

The mechanism is two lines: "User-Agent: __usage_ai-training" followed by "Disallow: /" opts out all content, and operators can also carve out specific directories. The token comes from an individual Internet-Draft in the Internet Engineering Task Force (IETF) AI Preferences working group, titled "Machine-Readable Training Preference Signal," authored by Anthropic standards lead Kevin Kelley and dated October 30, 2026, expiring on May 3, 2027. It is only an individual draft: it is not a formal IETF standard and carries no endorsement from the IETF.

Uneven rollout: Anthropic first, others in 2027

The four companies are moving at different speeds. Anthropic has already begun implementing the token; Microsoft targets February 1, 2027; Google says it will support it during 2027; and OpenAI says it aims to implement the signal in the coming months and will share implementation details later. OpenAI adds that the new signal will not replace its existing robots.txt crawler controls, which continue to apply. Microsoft says it will offer more granular controls, letting publishers opt out of AI training and AI grounding (live retrieval into chat answers) separately; Google says that a site using the shared token without naming Google-Extended will be treated as having disallowed Google-Extended.

Training only, and voluntary

The signal has two important limits. First, it covers training only: it does not stop a crawler from visiting or caching a page, and it does not affect search indexing or AI answers that quote pages. Second, it is voluntary, with no enforcement mechanism; and updating robots.txt does not remove content already collected into datasets or trained into models — it only affects future training runs. In short, this is a coordinated attempt to make the opt-out choice simpler, not a one-click erasure.

This report is compiled from OpenAI's announcement, Microsoft's Bing blog and related reporting; each company's exact launch date and implementation details remain subject to its later announcements.