JP — ITmedia
OpenAI, Google and Microsoft to Honor Common robots.txt Signal for Opting Out of AI Training
Policy2 min read
OpenAI, Google, and Microsoft said they will support one shared way for website owners to tell AI companies whether their content can be used for AI training. The shared signal is written in the robots.txt file that websites already use to control web crawlers.
What happened
On October 30 (US time), OpenAI, Google, and Microsoft each said they will support a common system. It lets website owners state their wishes about AI training to many AI companies at once. Anthropic is also building it. The idea was proposed in a draft document sent to the standards group IETF by Kevin Kelly, a standards lead at Anthropic.
Why this is needed
Until now, website owners had to find each AI company's crawler (a program that visits web pages and copies them) and write separate rules for each one. The new system uses one shared signal in robots.txt. A website owner writes it once, and each AI company that supports it can read it.
Key facts
- The shared signal uses the policy token "__usage_ai-training" under the "User-Agent" line in robots.txt.
- "Disallow" means the site does not allow AI training. "Allow" means it does.
- Example: "Disallow: /" with "Allow: /press-kit/" means no AI training on any page except /press-kit/.
- If the shared token is not set, the site is treated as having no stated wish about AI training.
- The signal does not block crawling. Crawl rules for each crawler still apply.
Timelines by company
- OpenAI: aims to support it within the next few months.
- Google: aims for 2027 with other AI companies. Google-Extended is still available.
- Microsoft: starts reading the new settings on February 1, 2027. Old NoArchive and NoCache settings must move to the new ones by March 1, 2027.
- Anthropic: has started building support, but full support across its services will take time.
Why it matters
Many publishers have sued AI companies over how their content is used, so website owners want clear control. A shared signal could cut the work of managing many separate rules. It also matches the EU AI Act's code of practice for general-purpose AI, which accepts robots.txt as a main way to express machine-readable wishes. However, companies can still set their own tools and private deals that override the shared signal, so the rules are not yet the same everywhere.
Glossary
- robots.txt
- A simple text file on a website that tells web crawlers which pages they may visit.
- IETF
- The Internet Engineering Task Force, a group that writes internet standards.
- Google-Extended
- Google's own setting that lets site owners control whether their content is used for Google's AI training.
- Anthropic
- An AI company that makes the Claude AI models.