OpenAI cannot rule out critical cyber capability in Astra evaluation
OpenAI said preliminary internal evaluations of its unreleased Astra model showed enough agentic coding and cyber performance that it could not rule out its Critical threshold, prompting stricter isolation, universal risky-action monitoring, and pauses on work that lacked upgraded controls.
0 comments · 0 votes
Sign in to join the discussion →
No comments yet. Start the discussion.
Why it moved the index
A possible critical cyber threshold would directly expand consequential autonomous attack capability. Magnitude 62 reflects the zero-day and end-to-end hardened-target stakes, while confidence 55 is capped because OpenAI's evidence is preliminary, internal, and says only that Critical capability cannot yet be ruled out.
Assessment history
-
R1
Toward 62 · confidence 55
Initial inclusion from OpenAI's dated disclosure of a completed preliminary capability assessment and resulting control changes.
11 Aug 2026
Share this page
-
DoomBench assesses “OpenAI cannot rule out critical cyber capability in Astra evaluation” as evidence moving toward doom, with magnitude 62 and confidence 55 out of 100 in the capability gains category.
-
The DoomBench assessment of “OpenAI cannot rule out critical cyber capability in Astra evaluation” is based on reporting from OpenAI and records the editorial rationale, source quality, attribution, and revision history.
-
DoomBench summarizes “OpenAI cannot rule out critical cyber capability in Astra evaluation” as follows: OpenAI said preliminary internal evaluations of its unreleased Astra model showed enough agentic coding and cyber performance that...
https://www.doombench.com/news/openai-cannot-rule-out-critical-cyber-capability-in-astra-evaluation-2026-08-07