OpenAI Delays Astra Release to Bolster Cybersecurity Safeguards
The story in brief
OpenAI delayed Astra’s release to enhance cybersecurity, calling it the first model meeting its 'Critical cybersecurity capability threshold.' Astra identifies and develops zero-day exploits in hardened systems autonomously, rejecting 91.5% of harmful requests. It wasn’t involved in the Hugging Face incident, but lessons from that event shaped its safety strategy. Advanced features will initially be limited to testers.