Anthropic, Google, Microsoft and OpenAI Agree to Support an AI Training Opt-Out Token
TLDR
Anthropic, Google, Microsoft and OpenAI have agreed to support a proposed 'policy token' that lets publishers opt out of AI model training with a few lines of text.
Optimixed Overview
Optimixed's original report on this story.
Publishers are pushing for more control over how AI companies scrape and use their content, and Digiday reports that Anthropic, Google, Microsoft and OpenAI have agreed to support a new mechanism for opting out of AI training. The proposed standard, called a policy token, is described as a simple few-line text addition, though the article notes the underlying problem is more complicated than it first appears.
How we got here: earlier coverage
- Nov 5EU Court Adviser Proposes Three Conditions for AI Training in Google Case
An EU court adviser in a Google case has backed AI training on three conditions: honouring robots.txt opt-outs, skipping pirate sites, and proving anti-memorisation steps.
- Nov 1Google, Microsoft and OpenAI Plan to Honor a Shared robots.txt Line for Opting Out of AI Training
A shared robots.txt line could let sites opt out of AI training with several providers at once, but the companies involved are handling it differently.
- Oct 31Sites Using NoArchive or NoCache Lose Microsoft's AI Opt-Out on March 1, 2027 Unless They Switch
Sites using NoArchive or NoCache will lose their Microsoft AI opt-out on March 1, 2027 unless they switch, and Google's token will also cover Gemini grounding.
- Oct 31Google and Bing will honor a shared robots.txt signal to block AI training
Google and Bing said they will honor a shared robots.txt signal that blocks AI training, and a study put AI Overviews click-through at 0.9%.