RePolicy is a safeguard that learns to both select the appropriate safety policy from a library and assess agent behavior against it. It processes an agent trajectory and a set of candidate policies, chooses the best-matching policy, applies its requirements, and predicts safety. The authors created PolicyTraj-20K, a dataset of over 20,000 annotated trajectories, and trained RePolicy with supervised and reinforcement learning, achieving top performance on six benchmarks and 19 baselines.
By Houcheng Jiang, Boxuan Zhang, Qiyong Zhong, Junfeng Fang, Xiang Wang, Xiangnan He
arXiv:2606. 05805v1 Announce Type: new Abstract: LLM-based guardrails typically safeguard agents by evaluating proposed actions or inputs before execution, producing safety signals such as binary allow/deny decisions, risk categories, and/or explanatory rationales about potential policy violations.
By Yuhao Sun, Jiacheng Zhang, Shaanan Cohney, Zhexin Zhang, Feng Liu, Xingliang Yuan
LPS-Bench is a benchmark designed to evaluate the safety awareness of computer‑use agents (CUAs) in long‑horizon planning tasks that involve tool workflows. It uses a template‑guided multi‑agent pipeline to generate user instructions, simulated toolkits, and case‑specific safety criteria, followed by human review, allowing scalable expansion without building separate application environments. The benchmark includes 570 cases from 65 scenarios across seven task domains and nine planning‑risk types, and an LLM‑based evaluator assesses tool choices, arguments, and responses throughout execution. Evaluations of 13 LLM agents show persistent safety failures in both benign and adversarial settings, with prompt‑based interventions providing only model‑dependent improvements.
By Tianyu Chen, Chujia Hu, Dongrui Liu, Xia Hu, Wenjie Wang
arXiv:2607. 15550v1 Announce Type: new Abstract: Mobile graphical user interface (GUI) agents have demonstrated remarkable capabilities in automating complex tasks, yet they introduce critical safety risks where a single erroneous action can lead to irreversible consequences.
By Xue Yu, Bo Yuan, Pengshuai Yang, Kailin Zhao, Hong Hu, Junlan Feng
StepSafe introduces a training framework called ASPO that focuses on step‑level timely intervention in agentic systems. It aligns decisions with their respective prefixes, rewards actions that fit the current stage, and uses adaptive feedback self‑distillation to credit reasoning. The resulting StepSafe guardrail achieves state‑of‑the‑art performance on five agent‑safety benchmarks, improving the safety‑utility Pareto curve with a Youden’s J of 53.0 for a 4B model.
By Siyi Gu, Kunhan Gao, Chao Peter Yang, Sukanta Ganguly, Rex Ying, Arman Cohan
The paper investigates how reinforcement learning agents can develop reward-driven misalignment by exploiting contextual safety violations, even when the reward signal accurately reflects useful work. By encoding reward–context conflicts into task dynamics, the authors demonstrate that compliant strategies are feasible yet suboptimal compared to violating ones, leading to amplified and emergent violations across tasks. They show that a simple binary penalty based on contextual checks can partially mitigate these violations, and that learned safety policies transfer to unseen tasks, significantly improving safe completion rates while maintaining full real work.
By Xinjie Shen