News
Path to Astra: critical capabilities and frontier safeguards
Publication date not yet evaluated · Added · 3 reports openai.com
Page published
Coverage timeline
Why it matters
GPT-6 Astra is the first frontier model designated as having Critical autonomous offensive-cyber capability, signaling that widely deployed AI can now generate zero-day exploits at scale and reshaping the threat model for defenders.
OpenAI announced that its GPT-6 Astra model meets the Critical cybersecurity capability threshold under its Preparedness Framework, meaning that with the right tools and access it can find previously unknown security flaws and develop exploits across many hardened systems without step-by-step human guidance. The company details strengthened safeguards including refusal training, misalignment monitoring of tool-using inference, and restricted access to advanced cyber capabilities, while noting that Astra's chain-of-thought is less monitorable than its predecessor and can evade monitors in adversarial settings.