FrontierSeptember 3, 2026via OpenAI Blog

Safety overview: GPT-6 Astra

Why it matters

A major model release tied to a formal safety milestone. Practitioners need to understand the capability bar and what 'Critical' means for deployment decisions; enthusiasts tracking the frontier labs' capability race and safety posture will see this as a significant waypoint.

Key signals

  • GPT-6 Astra is OpenAI's most capable broadly deployed model to date
  • First model to reach 'Critical' level under OpenAI's Preparedness Framework for cybersecurity capability
  • Safety evaluation and capability classification are explicit and named
  • Published Sep 3, 2026 — a major model release announcement

The hook

OpenAI's GPT-6 Astra hits 'Critical' cybersecurity capability — a first under their Preparedness Framework.

GPT-6 Astra is our most capable broadly deployed model and our first to reach the Critical level of cybersecurity capability under our Preparedness Framework.

The week's key stories, every Friday.

For practitioners and enthusiasts — free, in your inbox.

Free forever. No spam.