Safety overview: GPT-6 Astra
OpenAI has officially released GPT-6 Astra as its most capable broadly deployed model. This model is the first to achieve the 'Critical' cybersecurity capability rating under the company's Preparedness Framework.
Verified State Diff
Impact & Verification Analysis
Enterprise users, cybersecurity researchers, and developers integrating high-capability models into security-sensitive workflows.
This release establishes a new benchmark for model capability and safety governance, signaling that GPT-6 has reached a level of proficiency that necessitates heightened oversight and specific security-focused deployment protocols.
Full Fact Overview
The announcement marks the transition of the GPT-6 architecture into a broad deployment phase, specifically highlighting the 'Astra' iteration. The 'Critical' cybersecurity rating indicates that the model has passed rigorous internal safety evaluations regarding its ability to assist in cyber-offensive operations, such as vulnerability research or exploit generation, as defined by OpenAI's internal Preparedness Framework. This classification suggests a shift in how OpenAI manages the release of high-capability models, prioritizing safety-gated deployment for models that cross specific risk thresholds.