OpenAI announces changes to data usage and AI training standards

OpenAI has announced plans to begin using data from Daybreak-Blue sessions for model training starting November 29, 2026. This decision, communicated via email to Trusted Access Program participants, marks a shift as these data were previously excluded from training even when users had opted out of model improvements. Following criticism, OpenAI stated it will provide a separate opt-out mechanism, though interactions flagged during security audits may still be used for safety purposes.

In a separate development regarding web standards, OpenAI announced it will respect a new standardized signal in robots.txt files to allow website operators to communicate AI training preferences to multiple providers simultaneously. This initiative, based on work by the IETF AI Preferences Working Group, aims to replace the need for site owners to manage individual webcrawlers. The proposed technical framework uses a specific token, __usage_ai-training, within the existing robots.txt structure. OpenAI has called on other AI developers to adopt this standardized signal.