OpenAI designates Astra Critical for cybersecurity
OpenAI says its forthcoming Astra model is the first to meet the Critical cybersecurity threshold under its Preparedness Framework, triggering delayed development, paused training runs and gated access at launch.
In a post published ahead of the model's release, OpenAI said Astra can find previously unknown security flaws and build working exploits against many well-protected systems without a person guiding each step — the first model the company has placed at the Critical level under its Preparedness Framework.
The company cited both automated and expert-led evidence. Astra scored a perfect 100% on ExploitBench. Citing contamination concerns, OpenAI built an internal set — "ExploitBench - Internal Port (June–August 2026)" — of 20 recently disclosed high-severity V8 vulnerabilities, on which Astra reached far higher arbitrary code-execution rates than GPT‑5.6 Sol using far fewer output tokens. During that evaluation the model discovered and used two zero-days as part of an exploit chain; OpenAI says it is disclosing them to maintainers. In expert assessments, Astra built a full browser-compromise chain that escaped the sandbox and ran commands on the host from an opened HTML file, and chained multiple flaws in a hardened OS into a local privilege escalation from unprivileged user to root.
OpenAI says it delayed parts of Astra's development and release while hardening safeguards, paused certain frontier training for two weeks after the OpenAI–Hugging Face incident, and restarted the previously paused large frontier RL run on August 28th. Some smaller experimental runs remain held back. Astra was not involved in that incident.
Advanced cybersecurity capabilities will initially go to a limited group of testers, with access through Daybreak Blue to follow for defensive use. No release date, pricing, or full system card has been published; OpenAI says details will come at launch.