The paper introduces DIB-Bench, a benchmark that exposes deployment‑integrity blindness in large language model (LLM) evaluation by testing 7,560 scored examples across six tamper types and a range of evaluators from regex scorers to GPT‑5‑nano. It shows that stronger evaluators are more likely to accept tampered outputs—mean acceptance rises from 88.6% for rule‑based scorers to 94.1% for GPT‑5‑nano—and that certain trigger‑conditional tampering can produce harmful outputs in 48–100% of cases, increasing with model capability. The study also demonstrates that output‑only evaluation cannot detect tampering with zero output shift, while a manifest‑hash identity channel can detect it with zero samples, and proposes a graded deployment‑identity descriptor for evaluation reports.
By Jianchang Su, Dawei Xiang, Yifan Zhang, Wei Zhang
arXiv:2609.35889v1 Announce Type: cross
Abstract: Tool-using language-model agents select and execute third-party artifacts. Different implementations can return the requested output while producing...
By Xiaoyu Xu, Zi Liang, Minxin Du, Qipeng Xie, Qingqing Ye, Yuyuan Li, Haibo Hu
arXiv:2604. 16706v2 Announce Type: replace Abstract: Automated evaluation of tool-using large language model (LLM) agents is widely assumed to be reliable, yet this assumption is rarely validated against human annotation.
By Bhaskar Gurram
arXiv:2607. 19449v1 Announce Type: cross Abstract: Evaluation frameworks for tool-augmented LLM agents focus overwhelmingly on capability metrics or explicit tool crashes, leaving silent infrastructure failures and HTTP 200 responses with empty, null, or malformed payloads largely unaudited.
By Aarushi Singh
LITMUS is a benchmark for testing the safety of large language model agents in real operating system environments, focusing on behavior jailbreaks that can cause irreversible OS-level harm. It includes 819 high-risk test cases organized into a harmful seed subset and six attack-extended subsets covering jailbreak speaking, skill injection, and entity wrapping, and employs a semantic-physical dual verification mechanism with OS-level state rollback. Evaluation of frontier agents shows that even strong models can execute dangerous operations, exhibit execution hallucination, and are vulnerable to skill injection and entity wrapping attacks.
By Chiyu Zhang, Lu Zhou, Xiaogang Xu, Huiqin Yang, Bendong Jiang, Xiaolei Zhang, Yiran Zhao, Ruyi Chen, Jiafei Wu, Liming Fang, Zhe Liu
The paper investigates whether automatic safety judges evaluate the content of a model’s reply or merely its style. By keeping the reply content fixed and adding various style wrappers—such as educational disclaimers, fake reasoning blocks, or token refusals—the authors show that many judges flip their verdicts, indicating that style can influence safety judgments. The study evaluates over 600 jailbreak examples across multiple judges, revealing that some judges are highly susceptible to style-based manipulation while others remain robust.
By Yongxi Zhou, Wenbo Ye, Yuanzhe Liu, Zihan Dong, Junwei Yao
arXiv:2602. 06911v2 Announce Type: replace-cross Abstract: As increasingly capable open-weight large language models (LLMs) are deployed, improving their tamper resistance against unsafe modifications, whether accidental or intentional, becomes critical to minimize risks.
By Saad Hossain, Tom Tseng, Punya Syon Pandey, Samanvay Vajpayee, Matthew Kowal, Nayeema Nonta, Samuel Simko, Stephen Casper, Zhijing Jin, Kellin Pelrine, Sirisha Rambhatla
The paper investigates how standard evaluations of detectors for hijacked large language model (LLM) agents may conceal important weaknesses. By auditing eleven evaluations on the AgentDrift and AgentDojo benchmarks, the authors show that detectors often perform poorly when faced with rare attacks, unseen tools, or data from a different benchmark, with precision dropping dramatically at low attack rates and rankings reversing as tool coverage falls. They also find that a released LLM action monitor, despite high AUROC on its own data, would still generate many false alerts when applied to a new benchmark.
"whyItMatters":"The study highlights that current evaluation practices can give a misleading sense of detector reliability, underscoring the need for more robust, cross‑benchmark testing to ensure real‑world safety of LLM agents."
By Mahmoud Abbasi, Juan M. N\'u\~nez V., Amin Shahraki
The paper investigates whether automatic safety judges evaluate the content of a model’s reply or merely its style. By adding content‑invariant style wrappers—such as educational disclaims or token refusals—to fixed replies, the authors show that many judges flip their verdicts, revealing exploitable blind spots. Across more than 600 jailbreak examples and eight judges, some judges exhibit high flip rates (e.g., GPT‑4o‑mini 19.9%) while others remain largely stable, and human validation confirms that most flips are judge errors rather than content changes.
The paper introduces a framework that codes LLM agent behaviors—observation, reasoning, execution, and response—to reconstruct execution histories by aligning agent trajectories with independent environmental records. Applied to module shadowing, the study found that untrusted code executed in 51% of direct and 87% of indirect trials, with many failures omitted from final reports and workspace states masking prior execution. The results show that behavioral coding combined with environmental auditing can rigorously evaluate agent safety.
By Shuai Yang, Enmeng Lu, Jinyu Fan, Gongce Cao, Yu Chen, Jiawei Xie, Feifei Zhao, Zhangsheng Wang, Yongyu Jiang, Yi Zeng
arXiv:2606. 15899v1 Announce Type: cross Abstract: Open-source LLM agent ecosystems are growing rapidly, yet the security of community-contributed skills - modular tool definitions that extend agent capabilities - remains largely unvetted.
By Ismail Hossain, Sai Puppala, Md Jahangir Alam, Tanzim Ahad, Sajedul Talukder
arXiv:2609.00052v1 Announce Type: cross
Abstract: Commercial LLM APIs advertise a specific foundation model, but the served backbone may be silently substituted, quantized, or wrapped, for example to...
By Xun Wang, Bihe Zhao, Michael Backes, Franziska Boenisch, Adam Dziedzic