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.
Sign in to join the discussion →
No comments yet. Start the discussion.
This is OpenAI's first confirmed Critical cyber-capability designation, replacing its earlier uncertainty with direct evaluation evidence that Astra can discover unknown flaws and develop exploits across hardened systems with little human guidance. The internal evaluations include two V8 zero-days, a browser-sandbox escape, host command execution, and privilege escalation. The evidence is controlled and does not show a real-world compromise. Restricted access, a 91.5 percent cyber-refusal result, chain-of-thought monitoring, and no unauthorized-action attempts in the reported honeypot simulations materially reduce deployment and control risk, but do not erase the step change in dual-use capability.
Reviewed videos from Artificial Intelligence Videos. These links add context and never change a DoomBench score.
Adds OpenAI's dated confirmation of Critical cyber capability and the safeguards applied before Astra release.
02 Sept 2026DoomBench assesses “OpenAI confirms Astra reaches Critical cyber capability and restricts access” as evidence moving toward doom, with magnitude 79 and confidence 98 out of 100 in the misuse and incidents category.
The DoomBench assessment of “OpenAI confirms Astra reaches Critical cyber capability and restricts access” is based on reporting from OpenAI and records the editorial rationale, source quality, attribution, and revision history.
DoomBench summarizes “OpenAI confirms Astra reaches Critical cyber capability and restricts access” as follows: OpenAI confirmed that its pre-release Astra model meets the Critical cybersecurity threshold after finding and exploiting...
