OpenAI releases its first model at the critical cyber level
OpenAI released GPT-6 Astra on Thursday. Daybreak cybersecurity customers received access first. Paid ChatGPT plans and the API are due over the following week. OpenAI classified Astra at the critical cybersecurity level under its Preparedness Framework. The company says the model can find unknown flaws and build working exploit chains against hardened systems. In its internal tests, Astra used two previously unknown flaws as part of an exploit chain.
Verified 8:18 AM PDT · 3 original sources
The evidence
What the reporting establishes
What happened
OpenAI released GPT-6 Astra on Thursday. Daybreak cybersecurity customers received access first. Paid ChatGPT plans and the API are due over the following week. OpenAI classified Astra at the critical cybersecurity level under its Preparedness Framework. The company says the model can find unknown flaws and build working exploit chains against hardened systems. In its internal tests, Astra used two previously unknown flaws as part of an exploit chain.
Pressure point
The company produced most of the capability and safety evidence. Its system card says advanced cyber access will remain restricted and that automated monitoring can pause legitimate work. TechCrunch also reported that Astra's reasoning can be harder to inspect as it uses fewer visible language tokens.
What to watch
Watch independent tests of Astra's cyber capability and safeguards. The practical test is whether OpenAI can give defenders useful access without letting misuse through or pausing routine work too often.
Audit the story
Original sources
Company claims remain company claims. Follow the reporting and judge the evidence directly.
Continue the morning