OpenAI, Google, Microsoft and Anthropic pledged one shared opt-out signal for AI training
Microsoft starts on 1 February 2027 and Google sometime in 2027, so the signal will not fully work for months.
In brief
4 AI giants, one robots.txt opt-out
OpenAI, Google, Microsoft and Anthropic will honor one signal to refuse AI training. Microsoft starts 1 February 2027. Crawling stays allowed, and no one enforces it.
Four of the biggest AI companies said on Friday, 30 October 2026, that they will honor one shared signal that lets websites refuse AI training. The companies are OpenAI, Google, Microsoft and Anthropic. The news comes from AFP wire copy and TechBuzz, plus an IETF draft that describes the signal.
How it works
Microsoft says the setting covers only future training, not crawling.
Publishers now have to write separate blocking rules for each bot that crawls the web. The new idea is a single line in a site's robots.txt file, the public note that tells bots what they may do. Any company that adopts the protocol would honor it.
The token is called __usage_ai-training. Anthropic's Kevin Kelley drafted it as an individual IETF Internet-Draft dated 30 October 2026.
Who starts when
The rollout is uneven:
- Anthropic says it has already started.
- OpenAI says it will come "in the coming months".
- Microsoft begins on 1 February 2027.
- Google plans sometime in 2027.
What it does and does not do
Microsoft says the setting covers only future model training. It does not stop bots from crawling a site. Scraping stays allowed until a publisher changes the setting.
Microsoft and Google both said in separate blog posts that opting out could lower the chance a publisher's content shows up in AI apps and experiences. Microsoft also plans finer controls for AI grounding, which means using web content to answer questions. It says Bing rankings are not affected.
Why publishers care
Bots that scrape articles for training have surged, and some publishers have gone to court. The New York Times sued OpenAI and Microsoft in December 2023, and both invoke fair use, a copyright exception. Since July 2025, Cloudflare has been turning away AI bots automatically for newly added customer sites. A standard called Really Simple Licensing already lets publishers charge visiting bots.
Meta spokesperson Andy Stone wrote on X that a common signal is "an important step". Meta has not committed to honoring the token.
The limits
Reports differ. TechBuzz describes the four firms as having adopted the standard, while AFP calls it a proposal that will not fully take effect for months. The IETF draft is individual and has no formal IETF standing. Critics say there is no enforcement mechanism and that the token covers training only. The OpenAI, Microsoft and Google blog posts could not be checked directly, and most reports are AFP copy.