Definition
Plain language
Whether a system does something on its own, without being asked.
As stated in the literature
Distinguished from capability: capability asks whether a model can perform a behavior when instructed; propensity asks whether it exhibits it unprompted under a given incentive structure.
Also called: propensities
Why it matters: Separating propensity from capability matters because a model that *could* misbehave but rarely does poses different risks — and demands different safeguards — than one that does so eagerly.
For example, a model might have the ability to write convincing phishing emails but a low propensity to actually produce one unprompted.
Heard on the show
“And separately, do they have the *propensity* to do it?”Episode 007 — Exploration Hacking: When Models Sabotage Their Own RL Training