OpenAI announced that its Astra model is the first to meet the Critical cybersecurity capability threshold under the company's Preparedness Framework. According to OpenAI, Astra can identify previously unknown security flaws and develop exploits across well-protected systems without human guidance. Evaluations showed the model achieved a 100% score on ExploitBench and discovered two zero-day vulnerabilities during internal testing, alongside demonstrating advanced capabilities in escaping sandboxes and executing host commands. Due to these advanced capabilities, OpenAI delayed parts of Astra's development and release to strengthen protections against cyber misuse and unauthorized model actions. The company implemented safeguards including training the model to more reliably refuse harmful cyber requests, expanding monitoring systems, and applying a more conservative model-behavior boundary for higher-risk accounts. OpenAI plans to make Astra available soon, with advanced cybersecurity features initially restricted to a group of testers and Daybreak Blue access to support defensive use. Further details regarding safety, security, and alignment testing will be shared in the model's system card at launch.
Path to Astra: critical capabilities and frontier safeguards
OpenAI has designated its Astra model as meeting the Critical cybersecurity capability threshold under its Preparedness Framework. The model can identify unknown security flaws and develop exploits across well-protected systems without human guidance. OpenAI delayed parts of Astra's development to implement stronger safeguards.
Why it matters
Astra is the first OpenAI model to meet the Critical cybersecurity capability threshold under the Preparedness Framework, with stronger safeguards for release Reported by OpenAI on September 1, 2026.