OpenAI, Google, Microsoft and Anthropic Back a Common Opt-Out Signal for AI Training

By Dafa Attas

Photo: dedelacrapule / Pixabay

Four of the largest AI developers, OpenAI, Google, Microsoft and Anthropic, have said they will respect a single, common signal that website owners can use to say they do not want their content used to train AI models. The commitment was announced on 30 and 31 October 2026 and reported by RTÉ, Reuters (via CNA) and others; OpenAI also published its own announcement.

Until now, a publisher that wanted to keep its pages out of AI training usually had to block each company's crawler separately, often by naming different user agents in a robots.txt file. A shared signal is meant to replace that patchwork with one instruction that all participating companies read the same way. It is based on an IETF Internet-Draft written by Anthropic's Kevin Kelley: a "policy token" group for AI training inside the existing robots.txt file. The draft has no formal standing in the IETF standards process yet.

How the companies are rolling it out

According to the reports, the timing differs by company:

• Anthropic is already rolling the signal out.
• Microsoft plans to honour it from 1 February 2027.
• Google expects to support it during 2027.
• OpenAI says it will follow in the coming months.

Meta's spokesperson also expressed support, without saying whether Meta will honour the signal itself.

What the signal does not do

Two limits stand out. First, the signal only applies to future training. Content that has already been collected and used is not covered. Second, the default stays the same: unless a site owner actively sets the signal, scraping for training is still allowed. In other words, this is an opt-out system, not a permission system.

The copyright angle

For rights holders, a common signal makes refusal cheaper and easier to prove. A publisher that sets it can show a clear, machine-readable reservation of rights. That matters in places such as the European Union, where the text and data mining exception for commercial use only applies if the rights holder has not reserved its rights in an appropriate manner, such as machine-readable means for content made publicly available online (Article 4(3) of Directive (EU) 2019/790).

It does not settle the bigger questions. Whether training on copyright works without a licence is lawful in the first place is still being argued in courts and legislatures in many countries, and an opt-out tool offered voluntarily by AI companies is not the same as a legal right to be paid.

Why it matters

For news sites, blogs, stock libraries and other online publishers, the practical step is simple: follow the IETF draft and decide whether to switch the signal on once it is supported. For creators outside the EU, the value of the signal will depend on how national law treats a machine-readable reservation, which is still unsettled in most of Asia.

Sources

• RTÉ: AI companies to honour common opt-out signal for publishers
• CNA/Reuters: AI giants promise publishers new way to block some scraping
• OpenAI: Making it easier for websites to express AI training preferences (30 Oct 2026)
• IETF: draft-kelley-aipref-training-preference-signal
• EUR-Lex: Directive (EU) 2019/790 on copyright in the Digital Single Market, Article 4