Definition
An AI model's ability to perform tasks that are valuable for legitimate purposes—like finding software bugs or understanding biological processes—but that could equally be weaponised by a malicious actor. The same reasoning skill that helps a defender find a flaw faster also helps an attacker exploit it.
Why it matters
Regulators and security teams are increasingly treating frontier AI models as dual-use technology, similar to encryption or advanced semiconductors. This framing drives export controls, mandatory capability evaluations before release, and the first government orders to suspend access to specific models.