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.
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.
PASTABench introduces a benchmark of 1,139 multi-turn trajectories to evaluate proactive safety monitoring in large language models. It formalizes three dimensions of intervention—whether, when, and what risk—to address gaps in step-level isolation and post-hoc trajectory assessment. The study finds that proactive intervention is largely unsolved, with the best model achieving only 40.74% optimal-timing interventions, and reveals that smaller models’ safety scores are often driven by lexical overfitting rather than true risk comprehension.
SafeEvolve is an experience-driven framework that co‑evolves a harness and policy to align large‑language‑model agents with safety goals. It uses completed on‑policy trajectories to update safety prompts and hierarchical skills, then applies a two‑stage SFT‑RL training loop that bootstraps the policy with the evolved harness and refines it through verifier‑augmented rewards. Experiments on agentic safety benchmarks show that SafeEvolve improves the safety‑utility tradeoff, achieving a three‑fold reduction in ASR on AgentDojo for Qwen3.5‑4B while increasing benign utility from 59.79% to 61.86%.
The paper introduces SCOPE, a method that post‑trains computer‑use agents to balance task completion with safety by conditioning actions on environmental risk. It combines supervised fine‑tuning on three trajectory types—capability demonstrations, safe continuations, and explicit refusals—followed by reinforcement learning to improve performance. Experiments starting from Qwen3.5‑9B show that SCOPE‑RL achieves high task success and attack‑avoidance rates, outperforming other agents on OSWorld and OS‑BLIND benchmarks.
The paper introduces MENTOR, a multi‑agent framework that provides on‑policy pedagogical intervention to address zero‑reward collapse in reinforcement learning from verifiable rewards. MENTOR diagnoses failures in real time and generates targeted prompts to guide the learning agent, integrating seamlessly with existing policy optimization pipelines such as GRPO. Experiments on math and code generation tasks show that MENTOR significantly reduces collapse, achieving up to 78.3% relative gains on Llama‑3.2‑3B‑Instruct.
SafeEvolve is an experience-driven framework that co‑evolves a harness and policy to improve safety alignment for LLM‑based agents. It uses on‑policy trajectory safety evidence to update safety prompts and hierarchical skills, producing auditable harness artifacts. The policy is trained via a two‑stage SFT‑RL pipeline that bootstraps with the evolved harness and then refines behavior through verifier‑decomposed rewards, yielding a better safety‑utility tradeoff on benchmarks such as AgentDojo.
arXiv:2607. 19913v1 Announce Type: new Abstract: Agent safety is moving from content moderation toward preventing operational failures before tool-using agents act.
SafeCoEvo is a test‑time framework that co‑evolves safety harnesses and guards for large language model agents. It uses a short‑term S‑Harness to quickly externalize recent runtime experience into explicit safety knowledge, and a long‑term GuardVPO to internalize accumulated experience into parametric risk‑judgment capabilities. This dual adaptation improves safety and task success, reducing unsafe outcomes by 10.05% and increasing task success by 12.15% over the strongest baseline.
Agent safety is moving from content moderation toward preventing operational failures before tool-using agents act. We propose Janus, a foresight-oriented framework for long-horizon agent safety that trains guards to anticipate delayed risks from partial trajectories.
ReSI is a recursive safety improvement framework that iteratively applies diverse red‑teaming methods to identify vulnerabilities in AI models, develops training recipes, and selects the update with the greatest safety gain while retaining capability. Across four dense and mixture‑of‑experts models, ReSI matches or surpasses frontier models on both in‑distribution and out‑of‑distribution safety benchmarks, and outperforms alignment baselines on nearly all safety evaluations. The framework notably reduces the mean X‑Teaming attack success rate from 86.01% to 31.45%, outperforming GPT‑5.6‑Luna’s 56.69% and demonstrating stronger resilience to unseen attacks.
arXiv:2609.36201v1 Announce Type: cross Abstract: Computer-use agents (CUAs), while capable of completing computer tasks in everyday and professional workflows, can cause unintended harm even under b...
The paper introduces COLAGUARD, a guardrail model that embeds multi-step safety reasoning into a continuous latent space, allowing efficient hidden-state propagation during inference. Compared to existing methods, COLAGUARD achieves an 8.24‑point macro‑F1 improvement over Llama Guard 3 and matches the explicit reasoning baseline GuardReasoner, while delivering a 12.9× speedup and a 22.4× reduction in token usage across ten moderation settings and eight safety benchmarks.