Four AI firms back one training opt-out⁠.

Google, Microsoft, OpenAI and Anthropic will honour one shared opt-out from AI training, written into robots.txt, The Next Web reports¹. Today a site that wants to say no has to write separate rules for each company's crawler¹. Under the new signal, a single rule keeps a site's content out of generative AI training¹. SiliconANGLE reports the same four companies backing the plan².

The signal comes from a draft submitted to the Internet Engineering Task Force by Kevin Kelley of Anthropic¹. It adds a new policy token to robots.txt, the file sites already use to tell crawlers what they may fetch, and the site then lists the paths it does not want used for training¹. The draft's own example blocks a whole site except its press kit¹. The setting would not apply to models that perform only non-generative tasks².

The four are not moving at the same speed¹. OpenAI aims to support the signal in the coming months, and its existing crawler controls, such as GPTBot, will still apply¹. Google aims to support it in 2027, and its crawler documentation says early 2027¹. Microsoft will honour the new preferences from 1 February 2027¹. Anthropic has started work on it, and Kelley wrote that adding support across its services will take some time¹.

Microsoft is also adding a separate page setting for grounding, when an AI answer draws on a live web page¹. Blocking grounding will significantly reduce a site's visibility in AI answers, Microsoft said, though it does not change a page's place in Bing search results¹. It also said the new preferences do not establish licensing or payment arrangements¹.

The draft is an individual submission, and the IETF has not endorsed it, so it has no formal standing yet¹. It states a preference about how content is used, not an order about crawling, and private agreements between a site and an AI company can override it¹. Google said the proposal is consistent with the EU's code of practice for general-purpose AI under the AI Act¹. Meta supported the idea but has not committed to using it².

A simpler no, not a payday

If your website carries work you would rather not see inside someone else's model, product copy, guides or research, this turns saying so into one line instead of a list of crawler names that goes stale. Nothing changes today. The dates run into 2027, the standard is a draft, and a preference is not a lock. A crawler that chooses to ignore it is no more bound than it was yesterday.

None of this comes from a British body, and neither the ICO nor DSIT is part of it. Two points matter for a UK firm. Keeping your pages out of training is not the same as keeping them out of AI answers, and Microsoft is plain that switching off grounding makes you harder to find in its AI tools. And nobody is offering to pay. Decide which you care about more, being found or being kept out, before your web agency adds the line for you.

Also today

• Santander puts a figure on what AI earned it

  Santander said AI initiatives generated 194 million euros of business value in the first nine months of 2026, as more firms start to put numbers on AI returns³.

• Binance says it has covered 95 per cent of hack losses

  Binance says it bought replacement tokens with company money to cover 95 per cent of the $6.3bn stolen in a hack the FBI blames on North Korea⁴.

• Debt at data centre landlords feeds AI bubble fears

  Neocloud firms that lease computing power to AI developers are under scrutiny, with CoreWeave carrying $35.1bn of debt and its shares down a quarter in a month⁵.

• Anthropic's text watermark has a gap on two older models

  Anthropic is adding its text watermark to four more Claude models, but on Opus 4.6 and Sonnet 4.6 changing a single sampling setting is enough to avoid it⁶.

• AI spending pushes Amazon's free cash flow below zero

  Amazon's free cash flow fell to minus $42.4bn over the twelve months to September, mostly because of AI spending, while AWS revenue rose 39% to $46bn⁷.