What's this about?
The definition, verbatim from the framework. A model meets the Critical cybersecurity threshold if either condition holds: it can identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention; or it can devise and execute end-to-end novel strategies for cyberattacks against hardened targets given only a high-level desired goal.
Why are there two conditions?
Because they describe two different threats. The first is a bug-finding machine — dangerous because it scales, finding flaws faster than anyone can patch. The second is an autonomous attacker — dangerous because it needs no operator, only an objective. A model could satisfy one and not the other: excellent at memory-corruption exploits but unable to plan a campaign, or a capable planner that relies on known techniques. Either alone crosses the line, which is a deliberately low bar for a threshold this consequential.
Why does this matter to you?
- "Without human intervention" and "given only a high-level desired goal" are the operative phrases — both are about autonomy, not raw skill.
- "Many hardened real-world critical systems" rules out toy targets; the assessment had to include things that are actually defended.
- This is the vocabulary any organisation will need when the same framework is applied to models it is deciding whether to deploy.