The Preparedness Framework is OpenAI’s published policy for tracking and preparing for frontier AI capabilities that could create new risks of severe harm. OpenAI introduced an initial “beta” version in December 2023 and released version 2 on April 15, 2025, which streamlined and clarified the approach.
The framework names a small set of Tracked Categories where a model’s capabilities could enable serious harm - in version 2 these are biological and chemical capabilities, cybersecurity, and AI self-improvement - and adds Research Categories for emerging areas that do not yet meet the bar for full tracking. For tracked capabilities it defines two thresholds. A model at “High” capability could amplify existing pathways to severe harm and must have safeguards that sufficiently reduce that risk before it is deployed. A model at “Critical” capability could open unprecedented new pathways to harm and must have safeguards in place not only before deployment but during development.
This structure makes the framework a concrete “if-then” commitment: certain measured capabilities trigger certain mandatory safeguards, with an internal process for assessing models against the thresholds. It parallels Anthropic’s Responsible Scaling Policy and DeepMind’s Frontier Safety Framework, and together these documents form the backbone of voluntary frontier-safety governance among the leading labs.
For policymakers and enterprise customers, the Preparedness Framework matters because it puts a vendor’s safety commitments in writing and ties them to capability levels, giving an external reference point for what protections should exist before a powerful model ships.