
EU Security Teams Have No Say Over OpenAI's Astra
OpenAI confirmed on September 1, 2026 that its Astra model meets the Critical cybersecurity threshold under its own Preparedness Framework, the first model ever so designated. It found two real zero-days on its own. OpenAI alone decides who gets access.










