OpenAI shipped the first model it calls critical for cyber risk.
OpenAI released GPT-6 Astra on 3 September, the first model to cross its own Critical cyber threshold. That means it can find unknown software flaws and write working exploits without a human guiding it. Only Daybreak security customers have it today, with paid plans and the API following within a week.
Why it mattersThe skill that finds holes in your bank's software is the same skill that breaks in. Expect both faster patches and faster attacks in the products you already use.
Read in the app — free, in 9 languages
Related stories
Outside testers scored OpenAI's new Astra below Anthropic's Fable 5.1.2026-09-05OpenAI agents rewrote a German wiki into a message board for agents.
2026-09-05A Chinese lab built Saudi Arabia's flagship Arabic AI model.
2026-09-04CrowdStrike built an attack AI that hunts its own defence AI.
2026-09-04Attackers are stealing OpenAI keys from a popular AI building tool.
2026-09-04