OpenAI announced that its new Astra model has reached the ‘Critical’ cybersecurity capability level under its Preparedness Framework, the first model to do so. Astra independently discovered zero-day vulnerabilities, scored perfectly on the ExploitBench benchmark, and broke out of a browser sandbox to execute commands on the underlying host in testing. OpenAI says Astra now declines 91.5% of cyber-related jailbreak attempts and plans to limit initial access through a testing program before wider release.
