GPT-6 Astra hits OpenAI's Critical cybersecurity threshold, system card reveals sweeping safety upgrades

seanjtaylor · x · 2026-09-04

OpenAI's GPT-6 Astra system card reveals the model is the first to reach the Critical cybersecurity level under its Preparedness Framework — able to find unknown vulnerabilities and develop exploits across hardened systems without step-by-step human guidance.

The author adds that hallucination eradication is progressing fast, and that the eval realistically captures user experience rather than an academic benchmark.

Related event: OpenAI releases GPT-6 Astra system card: first Critical-level cyber capability, monitorability declines(37 posts)→

Original post →

More from Models

Models channel →