Metraipolis › archive › dc_fcd971ec38e2
SentryBench: A Cost-Sensitive Benchmark for Adversarial Robustness of Tool-Using LLM Agents in Public-Safety Deployments — arXiv:2610.34535
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.34535
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-27T08:01:29.928Z
- SHA-256 of the archived text
b536de7a2657d4e746e1ab1ed11818111a3a9f5a5f482b7fe2e5ad648ef3e98f- Stable URL of this page
https://metraipolis.com/documento/dc_fcd971ec38e2
How to cite
SentryBench: A Cost-Sensitive Benchmark for Adversarial Robustness of Tool-Using LLM Agents in Public-Safety Deployments — arXiv:2610.34535 — arXiv (en), https://arxiv.org/abs/2610.34535 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_fcd971ec38e2 (Metraipolis archive dc_fcd971ec38e2, sha256 b536de7a2657d4e746e1ab1ed11818111a3a9f5a5f482b7fe2e5ad648ef3e98f)
BibTeX
@misc{dcfcd971ec38e2,
title = {SentryBench: A Cost-Sensitive Benchmark for Adversarial Robustness of Tool-Using LLM Agents in Public-Safety Deployments — arXiv:2610.34535},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.34535},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_fcd971ec38e2; sha256 b536de7a2657d4e746e1ab1ed11818111a3a9f5a5f482b7fe2e5ad648ef3e98f},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.34535. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Large language model (LLM) agents are increasingly wired to consequential tools that send public alerts, dispatch responders, and query citizen records, where one erroneous tool call carries real-world cost. Existing agent-safety benchmarks measure generic harm or prompt-injection success and implicitly treat refusing as the safe default. In public-safety deployments that assumption breaks: an agent too cautious to dispatch help fails as surely as one that triggers a false alert. We present SentryBench, which scores agents on a two-sided, cost-sensitive frontier jointly measuring attack success, over-refusal (missing a real emergency), and over-reaction (unwarranted action) under adversarial tool use, with scenarios grounded in real emergency-call transcripts, disaster messages, and real injection/jailbreak payloads. We formalize deployment harm as an asymmetric cost functional, prove naive caution is harm-dominated whenever it refuses more genuine emergencies than the attack harm it averts, and prove a targeted trust gate adds zero over-refusal on clean input. Across 840 scored episodes over three frontier models and three defenses, refuse-more caution drives over-refusal to 100 percent on two of three models for at most a 3.7-point change in attack success and yields the worst harm everywhere, whereas our trust-gating wrapper dominates the frontier (lowest harm on every model, over-refusal at or below baseline, higher containment). Provider content filters also silently perform part of the defending and must be reported separately. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Cryptography and Security (cs.CR) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2610.34535 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-27T07:55:30.913Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).