CONCEPT
The Forged Sign Stimulus
The ethological concept, drawn from Konrad Lorenz and Niko Tinbergen, that a learned response is keyed to a small set of trigger features—not to a meaningful situation—and can therefore be reliably elicited by presenting the features without the situation, which is the precise structure of a jailbreak.
In
Konrad Lorenz and Niko Tinbergen’s ethology, a sign stimulus is the small set of trigger features that releases an innate response—the red belly that triggers a stickleback’s territorial attack, the egg shape that triggers a goose’s retrieval behavior. The lock (the innate releasing mechanism) does not inspect the locksmith: present the features without the situation and the response fires at nothing real. Tinbergen reported that male sticklebacks in tanks by his window would orient aggressively toward a red Royal Mail van passing in the street. The forged sign stimulus is the counterfeited key: a stimulus that presents the trigger features of a situation without that situation being present, causing the releasing mechanism to fire. Applied to
large language models, every jailbreak is a forged sign stimulus: the attacker presents trigger features (framing, assigned role, phrasing, context) that key the model’s compliance response without the situation that