AI Training Opt-Out: One robots.txt Line for Google, OpenAI, Anthropic, Microsoft (2026)
An AI training opt-out signal is a shared robots.txt rule that tells AI companies not to use a website's content to train generative models. If you run a company site or blog, you can now send that message to four AI leaders at once. On Oct. 30, Google, Microsoft, OpenAI and Anthropic said they will honor the same signal.
Why the AI Training Opt-Out Used to Be Tedious
Until now, site owners had to name each company's crawler in robots.txt (the file that tells bots which pages they may visit): GPTBot for OpenAI, Google-Extended for Google, ClaudeBot for Anthropic. Meanwhile, newspapers sued OpenAI and Microsoft over copyright and pushed for both a right to refuse and payment. The EU AI Act's code of practice also backs robots.txt as an opt-out channel.
How the __usage_ai-training Signal Works and When It Starts
Anthropic's Kevin Kelley filed the draft with the IETF (the internet's standards body) on Oct. 30. A site adds a "User-Agent: __usage_ai-training" group and lists the paths it wants excluded; "Disallow: /" covers the whole site. Anthropic has started work, OpenAI expects support in the coming months, Google targets early 2027, and Microsoft begins on Feb. 1, 2027.
Three Limits of the AI Training Opt-Out
First, the default is allow: silence counts as no preference. Second, it is not retroactive, so content already used for training stays in. Third, it has no teeth. The draft lacks IETF endorsement, private deals can override it, and Microsoft says it creates no "licensing or payment arrangements."
What It Means for Publishers and the AI Industry
Meta voiced support for a common signal but has not said it will comply, and Amazon and Apple are not among the four. Blocking also has a cost. Microsoft warns that blocking grounding (using live pages in AI answers) can significantly reduce a page's visibility in AI-generated answers. Web content and AI are moving from "scrape first" toward honoring a stated no, but bargaining over price has only begun.
Key Takeaways
① One shared line - One robots.txt group signals a training opt-out to all four.
② Rolling out by early 2027 - Anthropic has started, OpenAI within months, Google in early 2027, Microsoft on Feb. 1, 2027.
③ No teeth, no pay - It is not retroactive, not an official standard and not a payment deal.
The AI training opt-out hands every site owner a new question: how much content to give away. Next comes the harder fight over price.
👉 AI Music Copyright Exposed - 21 Million Songs in AI Training Data 2026 - also worth a read.
📌 Sources: The Next Web, Engadget, Microsoft Bing Blog, IETF, PPC Land (2026)



댓글
댓글 쓰기