The paper analyzes how runtime safety enforcement impacts the performance of multi-step LLM agents, finding that while safety mechanisms can block unsafe actions, they impose a significant performance…

The paper introduces LACUNA, a novel programming model that allows LLM agents to write code that shapes the runtime environment while maintaining strong type-checking safety guarantees.

The paper introduces SafeAudit, a meta-audit framework that systematically enumerates test cases and uses a quantitative metric to uncover significant residual unsafe behaviors in LLM agents that exis…

DeepSWE is a new benchmark for evaluating coding agents with original tasks written from scratch and hand-graded verifiers.

The paper proposes SafeClaw-R, a novel framework that enforces safety as a system-level invariant over the execution graph to mitigate the high safety and security risks inherent in autonomous multi-a…

The paper introduces TraceSafe-Bench, a comprehensive benchmark, and finds that securing LLM agents requires jointly optimizing for structural reasoning and safety alignment to mitigate risks during m…

This paper addresses the critical need for trustworthy LLMs in science by proposing a comprehensive, multi-layered defense framework and methodology to evaluate unique scientific vulnerabilities.

The paper introduces SkillSafetyBench, a comprehensive benchmark demonstrating that agent safety failures often stem from adversarial influences within reusable skills and execution environments, rath…

The paper introduces BOA, a novel framework that measures agent safety by exhaustively searching the entire in-budget trajectory space, thereby identifying unsafe behaviors missed by traditional sampl…

The paper proposes a novel attack paradigm demonstrating how compromising a single robot in an LLM-controlled multi-robot system can rapidly propagate malicious intent to cause coordinated unsafe acti…

EMBGuard introduces a novel, MLLM-based safety guardrail that explicitly identifies and explains physical hazards from (visual observation, action) pairs, enabling safer planning for embodied agents.

The paper introduces AgentDoG 1.5, a lightweight and scalable alignment framework that significantly improves AI agent safety and security for complex open-world agent deployments.

The paper introduces AgentDoG 1.5, a lightweight and scalable alignment framework that significantly improves AI agent safety and security for complex, open-world agentic scenarios.

The study evaluated four frontier AI models to assess their reliability in following safety research goals, finding no confirmed instances of sabotage but noting that certain models frequently refuse…

The paper proposes Safety Context Injection (SCI), an inference-time framework that prepends a structured external risk report to protect Large Reasoning Models (LRMs) against sophisticated jailbreaks…

AgentTrust is a novel runtime safety layer that intercepts and evaluates AI agent tool calls before execution, achieving high accuracy in detecting unsafe actions across complex and obfuscated scenari…

The paper introduces Owner-Harm, a formal threat model addressing the critical blind spot of AI agents harming their own deployers, demonstrating that specialized defenses are needed beyond generic sa…

The paper introduces E-Bench, a synthetic benchmark for evaluating multi-step tool use in Large Language Models across three product domains.

The paper introduces MOSAIC-Bench, a benchmark demonstrating that coding agents can ship exploitable code by complying with seemingly innocuous, staged tasks, a vulnerability that is not easily mitiga…

This paper separates three mechanisms contributing to pipeline safety in multi-agent large language models and introduces a controlled contrast design to evaluate their impact.