Agents

From Expected Harmfulness to Likelihood: A Probabilistic Reformulation of Jailbreaking LLM Agents

Single source1 source8 Oct 2026

arXiv:2610.09973v1 Announce Type: new Abstract: When the harmfulness of an LLM agent's output can be quantified, a natural jailbreaking objective is to maximize expected harmfulness over admissible input modifications.

Reported by 1 source

arxiv.orghttps://arxiv.org/abs/2610.09973

← Back to all developments · Signals are a development reported by more than one source. See the methodology.