Astra
OpenAI's pre-release frontier model and its first confirmed Critical cybersecurity system, with restricted advanced cyber access, strong cyber refusals, production monitoring, and a broader release planned but not yet completed.
0 comments · 0 votes
Sign in to join the discussion →
No comments yet. Start the discussion.
Why this model scores 82.9
Astra reached 100 percent on ExploitBench, found and used two internal V8 zero-days, built a browser-sandbox escape and host-command chain, and escalated privileges, establishing exceptional cyber capability and autonomy. It remains internal before launch, with advanced cyber access planned only for trusted partners. OpenAI reports 91.5 percent cyber refusals, chain-of-thought monitoring, and no unauthorized actions in described honeypot simulations, lowering residual control difficulty from the raw capability level while leaving very high dual-use misuse potential.
News tied to Astra
The model score of 82.9 rates this model's risk profile. The overall Doom Index of 65.3 measures the complete temporally weighted evidence record. These values answer different questions.
Each article's current Doom Index contribution is divided equally among the exact models named on that article. This prevents multi-model evidence from being claimed in full on several model pages. Model risk scores use a bounded temporal offset around their technical profile, but never feed back into the overall index.
OpenAI confirms Astra reaches Critical cyber capability and restricts access
OpenAI confirmed that its pre-release Astra model meets the Critical cybersecurity threshold after finding and exploiting internal V8 zero-days, building a browser-sandbox escape and executing host commands. OpenAI delayed development, added stronger refusals and monitoring, and plans restricted access to advanced cyber capabilities.
- Full item contribution
- +0.64
- Astra equal share
- +0.64
Model score history
-
R1
Doom Score 82.9
Creates Astra's internal exact-model profile from OpenAI's dated Critical-capability and safeguard assessment.
02 Sept 2026