Says Block, Still Acts: Why LLM Safety Judgments Fail to Govern Action in LLM Agents

Reported by arXiv cs.CR (Dongsheng Chen, Xiangyu Zhao, Xin Yao, Xuetao Wei)

Summary by Cyber Security News30 Sep 2026, 06:00 CEST

About this summary. This is a short, independently written summary of an article first published by arXiv cs.CR. Cyber Security News did not report or verify the underlying story. Read the original: https://arxiv.org/abs/2609.35870

arXiv:2609.35870v1 Announce Type: new Abstract: Large language model agents can correctly judge that an action should be blocked while still preferring to take it. We ask why this judgment-action disconnect arises, and whether explicit safety judgment causally governs subsequent action preference. Across three open-weight language models, safety-predictive information remains recoverable from action states, arguing against a simple information-loss account.

Read the full article at arXiv cs.CR →

Source attribution: headline and facts are from arXiv cs.CR (arxiv.org). Summary method: excerpt of the source description. See our source attribution policy.

Related coverage

SameFact: The Same Safety Facts Lead to Different Responses Across Interfaces

SINGED: Correct Outputs Do Not Certify Safe Execution in LLM Agents

SaplingGuard: A Multidimensional-Profile-Aware Multi-Agent Guardrail for Developmentally Safe Adolescent-LLM Interaction

Raising the Bar for Chinese Adolescent LLM Safety: A Culturally-Grounded, Fine-Grained Benchmark

Similarity Is Not Validity: Defending LLM Semantic Caches Against Poisoning

Cheap to Hypothesize, Costly to Verify: The Defense Surface of Agentic Vulnerability Discovery