OpenAI's Upcoming Astra Model Completes Voluntary U.S. Government Safety Review
OpenAI CEO Sam Altman confirmed that the company's upcoming Astra model has completed a voluntary pre-release safety review by the U.S. government. Astra is notably OpenAI's first model to reach the critical capability level for cybersecurity within its internal risk tracking framework. This milestone highlights increasing cooperation between leading AI labs and national security officials regarding high-risk frontier models. It also illustrates how voluntary oversight mechanisms are currently driving AI governance in the absence of comprehensive federal AI legislation. Under the U.S. government's voluntary framework, officials receive up to 30 days of early model access for safety evaluations, though the specific evaluation standards remain undisclosed to the public. Reports indicates that Astra possesses significant advanced capabilities in both cybersecurity access and mathematical reasoning.
## BACKGROUND
OpenAI utilizes an internal Preparedness Framework to track and manage severe systemic risks associated with frontier AI, categorizing risk capabilities into levels such as High and Critical. Models that trigger critical thresholds in domains like cybersecurity require heightened evaluation protocols and safety mitigations before deployment.