Google, OpenAI, Microsoft and Anthropic agreed on a single AI-training opt-out

Google, OpenAI, Microsoft and Anthropic announced on October 30 that they will voluntarily adopt a shared AI-training control for web publishers, built on the Internet Engineering Task Force's AI Preferences working group: instead of writing separate instructions for every AI developer, a website owner sets one preference in robots.txt and every participating provider reads it the same way.

The mechanism is a single policy token inside an ordinary robots.txt group. When a site uses the token to opt out of training and has not explicitly allowed Google-Extended, Google plans to treat that as if Google-Extended were disallowed; existing company-specific controls keep working alongside the shared signal, so owners can still tailor instructions per developer. Google says it will support the new control in 2027, and OpenAI says it will implement it in the coming months.

The catch is in the fine print: this is a draft proposal, not a standard, and robots.txt is a voluntary request rather than a lock — a scraper that does not cooperate can ignore it. What the labs actually agreed to is one shared opt-out form, not enforcement, and the test will be whether every provider honors the same token the same way once it ships.