SentryBench: A Cost-Sensitive Benchmark for Adversarial Robustness of Tool-Using LLM Agents in Public-Safety Deployments

Summary

SentryBench is a benchmark for LLM agents that call consequential public-safety tools, such as public alerts, responder dispatch and citizen-record queries. It scores attack success, over-refusal and over-reaction together with an asymmetric cost functional, using scenarios drawn from real emergency-call transcripts, disaster messages and injection/jailbreak payloads. Across 840 scored episodes on three frontier models and three defenses, refuse-more caution pushed over-refusal to 100 percent on two models for at most a 3.7-point change in attack success, while the trust-gating wrapper gave the lowest harm on every model.

Mitigation

The source proposes a targeted trust-gating wrapper as a defense, which it reports has the lowest harm on every model, over-refusal at or below baseline, and higher containment. It does not give a deployment fix beyond this defense. It also notes that provider content filters perform part of the defending and must be reported separately.

Topics

• Adversarial machine learning

Related items

• MediumIn-the-Wild Attack Abuses AI Agents' Trust in MarkdownSimilar attack · Dark Reading
• InfoI-HITL: Inalienable Human-in-the-LoopSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoThen there is prompt injection, a malicious email, ...: The Wild Wide West of OpenClaw Agentic System SecuritySimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• MediumMalicious downloads can trick AI coding agents into compromising themselvesSimilar attack · CSO Online
• InfoCoSA: The Conflict of Skills Attack in Real-World LLM Agent EcosystemsSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)