Nov 1, 2026

AI Giants Back One Training Opt-Out as Cursor Pulls Tampered Extension

Google, Microsoft, OpenAI and Anthropic will honour one shared robots.txt opt-out from AI training. Plus: a tampered Claude Code extension on Cursor's marketplace, and AI spending rattles markets.

Listen to this update

Four of the biggest names in artificial intelligence have agreed to read the same "do not train on my site" signal. Google, Microsoft, OpenAI and Anthropic said they will honour a single robots.txt rule letting publishers opt out of having their pages used to train AI models. Meanwhile, Cursor disclosed that tampered copies of a popular Claude Code extension were served through its marketplace for about 11 hours, and Big Tech's AI spending bill kept weighing on markets as October closed.

One rule instead of a dozen

Until now, a website that didn't want its content used for AI training had to find every AI company's crawler name and write a separate block for each one. The new proposal replaces that with a single shared label. Site owners add a group for a policy token called __usage_ai-training to robots.txt and list the paths they don't want used for training. They can also allow specific sections; the draft's own example blocks a whole site except its press kit.

The idea comes from an Internet-Draft written by Kevin Kelley, Anthropic's technical lead for standards, and posted to the IETF, the body that maintains many core internet standards. Google, OpenAI and Microsoft announced their support on the same day, according to TNW and Engadget.

The four companies are on different schedules:

Anthropic says it has started implementing the signal, though it expects edge cases to take time. The new rule sits alongside its existing crawler controls.
OpenAI says it will add support "in the coming months," and its existing GPTBot controls still apply.
Google plans support in early 2027. When a site uses the new token without mentioning Google-Extended, Google will treat it as a block on Google-Extended, the crawler that gathers material for Gemini training.
Microsoft will honour the signal from February 1, 2027. It replaces the NoArchive and NoCache tags Microsoft has used since 2023. Microsoft is also adding a page-level control over "grounding," the practice of pulling live web content into AI answers.

What the opt-out does not do

The new token states a preference about how content may be used. It doesn't tell crawlers to stay away. It won't remove content that has already been collected, and it won't stop pages from appearing in Google Search. A site that doesn't add the group hasn't stated any preference, and older rules that block all crawlers don't count as an opt-out. Private licensing deals between publishers and AI firms can also override it. Microsoft says the signal creates no licensing or payment arrangements.

The draft also has no formal standing yet. It is an individual submission, not an IETF-endorsed standard. Meta said on X that it supports a common signal but did not commit to honouring this one. Google says the approach fits the EU AI Act's Code of Practice, which treats robots.txt as a key way for publishers to state machine-readable preferences. The draft thanks staff from the Financial Times, BBC, Bertelsmann and the European Publishers Council.

Tampered Claude Code extension hit Cursor users

Cursor, the AI coding editor, published a security notice saying that versions 2.1.312 and 2.1.313 of the Anthropic Claude Code extension for Windows and macOS were malicious. They were available in Cursor's extension marketplace from about 2:27 a.m. to 1:35 p.m. Pacific time on October 29. Cursor says Anthropic did not build those versions, and it has no evidence that other versions or Anthropic's official distribution were affected.

According to Cursor, the code sent host details such as usernames, IP addresses and Git emails to an attacker's server. It also added an attacker-controlled SSH key, labelled svc-telemetry, to the user's authorized_keys file and tried to open a persistent reverse SSH tunnel. The step that turns on the SSH service needs admin rights, which the extension usually didn't have, so it likely failed on most machines. Even so, Cursor warns that uninstalling the extension may not be enough to clean an affected machine.

Cursor says it removed the versions, took its whole marketplace offline, checked that no other extensions had been tampered with, rotated its keys, rebuilt the marketplace and worked to take down the attacker's server. Its notice does not explain how the tampered package was published. It has released remediation scripts for both platforms and advises affected users to rotate credentials from a clean machine. BleepingComputer estimated that up to 5,000 developers may have been exposed.

AI spending keeps rattling markets

The backdrop to the week was money. Fortune estimates that Amazon, Alphabet, Microsoft, Meta and SpaceX together plan up to $828 billion in capital spending this year, mostly for AI. Four of those five companies reported negative free cash flow last quarter. TechCrunch reported that Alphabet's free cash flow was negative $9.1 billion, its second negative quarter in a row, and that Amazon's was negative $42.4 billion over the past 12 months. Microsoft was the only one of the four big cloud players whose cash coming in exceeded what it spent.

Financing the buildout is changing too. Fortune reports that Meta plans to issue new shares to raise up to $95 billion for AI infrastructure. Microsoft announced a $25 billion bond offering, its first in more than nine years. TechCrunch, citing S&P Global Market Intelligence, says the largest hyperscalers and Oracle have raised $326 billion in debt this year, about triple last year's total. On Friday the Nasdaq fell 2.9% and the S&P 500 fell 1.8%, their worst days since June. Nvidia fell 4% and Qualcomm 11%.