Definition
The difficulty of establishing who is responsible for an AI-enabled harm — whether a bad model, a malicious actor, or an autonomous system — when an incident occurs. This is a recognised gap across domains from biosecurity to cyber. Without reliable attribution, liability, deterrence and response all become harder.
Why it matters
If you can't reliably say who caused an AI-caused incident, you can't assign liability, deter bad actors, or respond credibly.