- OpenAI announced that its in-development model “Astra” is the first to reach the Critical cybersecurity threshold defined in its own Preparedness Framework.
- OpenAI says the model can discover unknown vulnerabilities and develop exploit code without human guidance.
- Before release, OpenAI is adding refusal training, stronger monitoring, and restricted access to the model’s advanced capabilities.
Why it matters
A leading AI lab has publicly confirmed the existence of AI capable of autonomously finding and exploiting zero-day vulnerabilities, pushing enterprises to urgently reassess the permissions granted to AI agents and their security monitoring.