Jailbreak Prompt
An attack technique in which a carefully constructed input prompt induces a model to bypass safety restrictions and produce prohibited content. It is a form of transient input manipulation.
Timeline (last 90 days)
Unlike traditional jailbreak prompts, Abliterated models work at the weight level: the restriction is not 'bypassed' but 'removed,' so prompt-layer defenses may fail completely.
Attackers may circumvent AI safety restrictions through jailbreak prompts, decomposing tasks step by step, or using the model for peripheral auxiliary work.
After one jailbreak prompt is patched, new jailbreak techniques quickly emerge.
All Facts (3)
Unlike traditional jailbreak prompts, Abliterated models work at the weight level: the restriction is not 'bypassed' but 'removed,' so prompt-layer defenses may fail completely.
50%UnverifiedAttackers may circumvent AI safety restrictions through jailbreak prompts, decomposing tasks step by step, or using the model for peripheral auxiliary work.
50%UnverifiedAfter one jailbreak prompt is patched, new jailbreak techniques quickly emerge.
50%