OpenAI’s Astra Achieves Critical Cybersecurity Milestone
- Astra is OpenAI’s first model to reach critical cybersecurity capabilities, able to independently identify unknown vulnerabilities.
- During evaluations, Astra discovered and exploited two zero-day vulnerabilities in the V8 engine as part of an exploit-chain attack.
- In tests, Astra refused 91.5% of dangerous requests, compared to only 59% for GPT-5.6 Sol.
- Astra achieved a perfect score on ExploitBench for developing exploits for known vulnerabilities.
- Access to Astra’s advanced capabilities will initially be limited to a small group of alpha testers before broader rollout.
OpenAI has enhanced safeguards around Astra due to risks of misuse by malicious actors and potential unauthorized actions by the model itself. The company plans a rapid-response program for newly identified vulnerabilities and continues rigorous testing protocols.
Astra’s ability to refuse dangerous requests at a rate of over nine times greater than its predecessor underscores its advanced security measures.(Source)
