OpenAI, Google, Microsoft And Anthropic Back Common AI Training Opt-Out Mechanism For Publishers
Last Updated: 2026-11-02 16:52:00
Major artificial intelligence companies, including OpenAI, Google, Microsoft and Anthropic, are backing a proposed mechanism that could make it easier for publishers and website owners to prevent their content from being used to train AI models.
The proposal introduces a shared system called a ‘policy token’, which would allow websites to communicate their AI content-use preferences through their existing robots.txt files. This could reduce the need for publishers to maintain separate instructions for every AI company.
Latest and Popular Mobiles
The development is significant for digital publishers, news organisations and independent website owners, many of whom have raised concerns about AI companies collecting and using online content to develop their models.
The proposal has been put forward through the Internet Engineering Task Force (IETF), with Kevin Kelley, a technical lead for standards at Anthropic, involved in the work.
Related Articles
Under the proposed mechanism, publishers would be able to add a specific policy token to their robots.txt files to indicate whether their content can be used for AI training.
At present, website owners often have to deal with different systems and instructions depending on the AI developer. A common mechanism could simplify this process by allowing participating companies to recognise a shared preference.
Google has announced plans to support the shared AI training control. The company said the mechanism would allow website owners to communicate their preferences across multiple AI providers through a single step.
Google also clarified that existing company-specific controls, including Google-Extended, would continue to work alongside the proposed mechanism.
The level of support from different AI companies is also becoming clearer. According to The Verge, Anthropic has already adopted the proposed signal, while Google and Microsoft plan to support it in 2027. OpenAI expects to implement it in the coming months. Meta has expressed support for a common standard but has not confirmed that it will honour the signal.
Microsoft is also considering separate controls for AI training and AI grounding, allowing publishers to distinguish between different ways their content could be used.
For publishers, the distinction is important. They may want to prevent their articles from being used to train AI models while still allowing their content to appear in search results or be referenced by AI-powered services.
The proposed system could offer a more consistent way to communicate these preferences, although its effectiveness will depend on how widely AI companies adopt and respect the mechanism.
It is also important to understand that the policy token is not a technical restriction that automatically prevents a website from being accessed or copied. Rather, it is a machine-readable signal expressing the website owner’s preferences. Its practical impact depends on whether participating companies recognise and follow those instructions.
The development comes amid growing tensions between publishers and AI companies over the use of online content. News organisations have increasingly questioned whether AI developers should be allowed to use their reporting, photographs and other material without appropriate permission or licensing arrangements.
For Indian publishers, the proposed standard could provide a simpler way to communicate content-use restrictions to international AI companies. However, website owners will still need to understand the available controls and monitor whether platforms comply with their stated preferences.