OpenAI, Google, Microsoft and Anthropic pledge a shared scraping opt-out for publishers

OpenAI, Google, Microsoft and Anthropic pledged on October 30, 2026 to give publishers and other websites a single standardized way to stop their content from training future AI models. Today, a publisher who wants to block AI crawlers has to set up a separate rule for each company's bot. The pledge replaces that with one signal every participating company recognizes and honors.

The rollout is staggered, and none of it is immediate. Microsoft plans to turn the signal on starting February 1, 2027. Google says sometime in 2027. OpenAI says "in the coming months." Anthropic says it has "already started." Meta did not join the pledge, but spokesperson Andy Stone wrote on social media that a shared signal reaching multiple companies at once is "an important step," and that Meta looks forward to working on the standard.

The companies named their own limits. The signal covers training only. It does not stop a company's crawler from visiting a site for any other reason, and scraping stays allowed by default until a publisher turns the setting on. Microsoft and Google both said in their own posts that a publisher who opts out may see less of its work surface inside AI apps and answers.

The pledge lands in the middle of a legal fight that has been running for nearly three years. The New York Times sued OpenAI and Microsoft in December 2023, alleging the companies trained ChatGPT on millions of its articles without a license or payment. OpenAI and Microsoft argue fair use covers the training, and no court has ruled on the claim yet. Cloudflare has blocked AI crawlers by default on new customer sites since July 2025, and a separate standard called Really Simple Licensing already lets publishers charge bots for access.

Read closely, the pledge answers a narrower question than the headline suggests. It settles whether a publisher can stop a model from learning from its site next year. It says nothing about the model that already exists. A reporter whose articles trained Claude or ChatGPT last year has no way to find out, today or after this signal rolls out, which of their sentences shaped which answer a reader got from either model yesterday. The opt-out governs consent going in. It is silent about where an answer came from going out, and that second question is the one a reader actually has when they are deciding whether to trust what a model told them.

That second question is the one ForIntel's reports are written to answer. Every claim in a ForIntel report traces back to a source the reader can click and check themselves, not because the law requires it yet, but because a reader deciding what to act on deserves to know where a claim came from. The scraping pledge may eventually settle what a model is allowed to learn from. It will not tell anyone, on its own, where any specific answer actually originated. That gap is still open, for every model this pledge covers, well after February 2027.