Incident·WS-0077·Feb 5, 2026·in the lab·Capability thresholds
OpenAI rates GPT-5.3-Codex its first 'High' cyber model
28/100
WS-2 ModerateHarm score 0, signal score 62. Autonomy: acted (the AI took a bounded action). How the scores work.
Harm 0/5
damage actually done
Reach 0/5
who was affected
Autonomy 2/4
how much the AI did on its own
Novelty 4/5
how new the behavior was
Permanence 0/5
how hard it is to undo
OpenAI's GPT-5.3-Codex system card says this is the first launch it treats as High capability in cybersecurity under its Preparedness Framework. OpenAI said it lacked definitive evidence that the threshold was met but could not rule it out, so it deployed the matching safeguards as a precaution.
Damage
none realized (capability threshold signal)
The capability behind it
- OpenAI's Daybreak puts cyber-tuned models in defenders' handsMay 12, 2026 · OpenAI · reliable for a defined task: partial
- Claude Mythos Preview reports thousands of critical software flaws; release withheldApr 7, 2026 · Anthropic · reliable for a defined task: partial
Sources
OpenAI, second source. The model is also treated as High in biology. It does not reach High for AI self-improvement.
Factor ratings and the score are this index’s judgments by a published formula; facts, dates and figures come from the linked sources. Where a lawsuit is pending, the summary reports allegations, not findings.
