OpenAI pauses Astra development after evaluations point to 'Critical' cyber capability
OpenAI suspended internal work on its unreleased Astra model after preliminary evaluations could not rule out the 'Critical' cybersecurity tier of its Preparedness Framework — a level defined as independently identifying and carrying out attacks on well-protected real-world systems. It is the first time OpenAI has flagged one of its own models at that tier; GPT-5.6 Sol topped out at 'High'. The company added universal monitoring and brought in government agencies and outside safety organizations for further testing, making pre-release capability disclosure an operating norm at frontier labs rather than a policy promise.
Source: techcrunch.com ↗
our preliminary evaluations indicate strong enough performance that we cannot rule out Critical capability level at this time
Why this matters
- → First model OpenAI flagged at 'Critical' cyber threat tier—threshold not reached before
- → Public pause signals frontier labs normalizing capability disclosure before release
- → Independent cyberattack capability on hardened systems raises autonomous agent risk