Anthropic, OpenAI, Google, and Microsoft Align on Shared AI Data Opt-Out Signal
Four major AI companies have agreed to support a common directive to allow publishers to opt out of AI training data scraping with a single robots.txt signal.
The rapidly expanding ecosystem of generative AI has brought the concept of copyright and data scraping into sharp focus. A major shift in how publishers can protect their content online has emerged as four of the world’s largest AI players—Anthropic, OpenAI, Google, and Microsoft—announce support for a unified way to signal their training datasets.
Currently, content creators must often navigate a fragmented landscape, blocking various crawlers individually to prevent their intellectual property from being used in large language models. However, the proposed solution streamlines this process, allowing for a single directive that publishers can implement within their existing robots.txt policies to opt out of AI training data collection.
This movement toward standardization was spearheaded by Anthropic engineer Kevin Kelley, who proposed the common directive through work with the Internet Engineering Task Force (IETF) as reported by Nieman Lab. The shared signal is designed to give publishers greater control over how AI models consume their data for training purposes.
Support for the new unified signal is currently staggered across the major platforms. Anthropic has already adopted the shared signal. OpenAI announced its commitment to implement the change in the coming months, while Google and Microsoft have set their deadlines further out, aiming for 2027.
For Google and Microsoft, the implementation involves specific technical directives. The companies will utilize a shared policy token named ’__usage_ai-training. Google has stated that it will process this specific token as a Google-Extended block, ensuring the signal is honored during its data collection process as detailed by Search Engine Journal. Meanwhile, Microsoft plans to begin honoring these new optional preferences starting February 1, 2027.
Furthermore, while Google has targeted implementation for early 2027, Microsoft is setting its rollout to February 1, 2027. This synchronized, albeit phased, commitment suggests a broadening consensus among the industry leaders that a standardized mechanism for data opt-outs is both necessary and achievable per Search Engine Journal. This effort moves the debate over AI training data from generalized ethical arguments toward specific, actionable technical implementations.