ClawSafety: "Safe" LLMs, Unsafe Agents

• URL: http://arxiv.org/abs/2604.01438v2
• Date: Sat, 04 Apr 2026 15:18:43 GMT
• Title: ClawSafety: "Safe" LLMs, Unsafe Agents
• Abstract summary: Personal AI agents like OpenClaw run with elevated privileges on users' local machines.<n>ClawSAFETY is a benchmark of 120 adversarial test scenarios organized along three dimensions (harm domain, attack vector, and harmful action type)<n>Each test case embeds adversarial content in one of three channels the agent encounters during normal work: workspace skill files, emails from trusted senders, and web pages.
• Score: 25.729388843970014
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Personal AI agents like OpenClaw run with elevated privileges on users' local machines, where a single successful prompt injection can leak credentials, redirect financial transactions, or destroy files. This threat goes well beyond conventional text-level jailbreaks, yet existing safety evaluations fall short: most test models in isolated chat settings, rely on synthetic environments, and do not account for how the agent framework itself shapes safety outcomes. We introduce CLAWSAFETY, a benchmark of 120 adversarial test scenarios organized along three dimensions (harm domain, attack vector, and harmful action type) and grounded in realistic, high-privilege professional workspaces spanning software engineering, finance, healthcare, law, and DevOps. Each test case embeds adversarial content in one of three channels the agent encounters during normal work: workspace skill files, emails from trusted senders, and web pages. We evaluate five frontier LLMs as agent backbones, running 2,520 sandboxed trials across all configurations. Attack success rates (ASR) range from 40\% to 75\% across models and vary sharply by injection vector, with skill instructions (highest trust) consistently more dangerous than email or web content. Action-trace analysis reveals that the strongest model maintains hard boundaries against credential forwarding and destructive actions, while weaker models permit both. Cross-scaffold experiments on three agent frameworks further demonstrate that safety is not determined by the backbone model alone but depends on the full deployment stack, calling for safety evaluation that treats model and framework as joint variables. Code and data will be available at: https://weibowen555.github.io/ClawSafety/.

Related papers

• Safety Testing LLM Agents at Scale: From Risk Discovery to Evidence-Grounded Verification [32.337075755765056]
  Vera is an endto-end automated safety testing framework that instantiates software engineering testing principles for non-deterministic agents.<n>We release Vera, comprising 1600 executable safety cases spanning 124 risk categories across execution settings.<n>These results indicate that modular, executable testing infrastructure is essential for rigorous and maintainable safety evaluation of rapidly evolving agentic systems.
  arXiv  Detail & Related papers  (2026-07-02T07:08:26Z)
• Understanding and Evaluating Claw-like Agent Security Through a Computer-Systems Lens [65.53494487819053]
  Claw-like AI agents (e.g., OpenClaw) are always-on processes with persistent access to credentials, files, tools, and external services.<n>We adopt a computer-system analogy: treating a Claw-like agent as an agentic computer system whose gateway runtime plays an OS-like mediation role.<n>We develop SafeClawArena, a benchmark of 406 adversarial tasks across four attack surfaces.
  arXiv  Detail & Related papers  (2026-06-29T18:00:45Z)
• AgentLens: Interpretable Safety Steering via Mechanistic Subspaces for Multi-Turn Coding Agent [53.82005364479556]
  Coding agents based on large language models (LLMs) demonstrate remarkable autonomous capabilities.<n>Existing safety mechanisms mainly rely on external guardrails.<n>We propose AgentLens, a white-box defense framework that performs runtime safety detection and representation-level mitigation.
  arXiv  Detail & Related papers  (2026-06-21T21:13:55Z)
• SafeClawBench: Separating Semantic, Audit-Evidence, and Sandbox Harm in Tool-Using LLM Agents [31.978790013349947]
  SafeClawBench is a benchmark for tool-using agent security with 600 controlled adversarial tasks.<n>We evaluate five agent endpoints under four prompt-level policies.<n>We find that these endpoints capture different failure modes.
  arXiv  Detail & Related papers  (2026-06-16T18:04:45Z)
• AgentTrap: Measuring Runtime Trust Failures in Third-Party Agent Skills [32.24433022603674]
  Third-party skills are becoming the package ecosystem for LLM agents.<n>Malicious skills do not need to ask the model to perform an obviously harmful action.<n>Instead, it can disguise the harmful behavior as part of a routine workflow.<n>We introduce AgentTrap, a benchmark for evaluating whether LLM agents can use third-party skills while resisting malicious runtime behavior.
  arXiv  Detail & Related papers  (2026-05-13T17:04:17Z)
• AgentTrust: Runtime Safety Evaluation and Interception for AI Agent Tool Use [2.9991161518367875]
  AgentTrust intercepts agent tool calls before execution and returns a structured verdict: allow, warn, block, or review.<n>We release a 300-scenario benchmark across six risk categories and an additional 630 independently constructed real-world adversarial scenarios.<n>On the 630-second benchmark, evaluated under a patched ruleset, AgentTrust achieves 96.7% verdict accuracy, including about 93% on shell-obfuscated payloads.
  arXiv  Detail & Related papers  (2026-05-06T11:38:16Z)
• Internal Safety Collapse in Frontier Large Language Models [65.00730294617382]
  This work identifies a critical failure mode in frontier large language models (LLMs)<n>Under certain task conditions, models enter a state in which they continuously generate harmful content while executing otherwise benign tasks.<n>We introduce TVD, a framework that triggers ISC through domain tasks where generating harmful content is the only valid completion.
  arXiv  Detail & Related papers  (2026-03-04T12:55:34Z)
• A Trajectory-Based Safety Audit of Clawdbot (OpenClaw) [77.1549110891026]
  We present a trajectory-centric evaluation of Clawdbot across six risk dimensions.<n>We log complete interaction trajectories (messages, actions, tool-call arguments/outputs) and assess safety using both an automated trajectory judge and human review.
  arXiv  Detail & Related papers  (2026-02-16T00:33:02Z)
• Breaking the Code: Security Assessment of AI Code Agents Through Systematic Jailbreaking Attacks [11.371490212283383]
  Code-capable large language model (LLM) agents are embedded into software engineering where they can read, write, and execute code.<n>We present JAWS-BENCH, a benchmark spanning three escalating workspaces that mirror attacker capability.<n>We find that under prompt-only conditions in JAWS-0, code agents accept 61% of attacks on average; 58% are harmful, 52% parse, and 27% run end-to-end.
  arXiv  Detail & Related papers  (2025-10-01T18:38:20Z)
• OpenAgentSafety: A Comprehensive Framework for Evaluating Real-World AI Agent Safety [58.201189860217724]
  We introduce OpenAgentSafety, a comprehensive framework for evaluating agent behavior across eight critical risk categories.<n>Unlike prior work, our framework evaluates agents that interact with real tools, including web browsers, code execution environments, file systems, bash shells, and messaging platforms.<n>It combines rule-based analysis with LLM-as-judge assessments to detect both overt and subtle unsafe behaviors.
  arXiv  Detail & Related papers  (2025-07-08T16:18:54Z)
• OS-Harm: A Benchmark for Measuring Safety of Computer Use Agents [60.78202583483591]
  We introduce OS-Harm, a new benchmark for measuring safety of computer use agents.<n> OS-Harm is built on top of the OSWorld environment and aims to test models across three categories of harm: deliberate user misuse, prompt injection attacks, and model misbehavior.<n>We evaluate computer use agents based on a range of frontier models and provide insights into their safety.
  arXiv  Detail & Related papers  (2025-06-17T17:59:31Z)
• DoomArena: A framework for Testing AI Agents Against Evolving Security Threats [81.73540246946015]
  We present DoomArena, a security evaluation framework for AI agents.<n>It is a plug-in framework and integrates easily into realistic agentic frameworks.<n>It is modular and decouples the development of attacks from details of the environment in which the agent is deployed.
  arXiv  Detail & Related papers  (2025-04-18T20:36:10Z)
• Multi-Agent Systems Execute Arbitrary Malicious Code [7.429202753218103]
  We show that adversarial content can hijack control and communication within the system to invoke unsafe agents and functionalities.<n>This results in a complete security breach, up to execution of arbitrary malicious code on the user's device.<n>We hope that these results will motivate development of trust and security models for multi-agent systems.
  arXiv  Detail & Related papers  (2025-03-15T16:16:08Z)
• SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents [58.65256663334316]
  We present SafeAgentBench -- the first benchmark for safety-aware task planning of embodied LLM agents in interactive simulation environments.<n>SafeAgentBench includes: (1) an executable, diverse, and high-quality dataset of 750 tasks, rigorously curated to cover 10 potential hazards and 3 task types; (2) SafeAgentEnv, a universal embodied environment with a low-level controller, supporting multi-agent execution with 17 high-level actions for 9 state-of-the-art baselines; and (3) reliable evaluation methods from both execution and semantic perspectives.
  arXiv  Detail & Related papers  (2024-12-17T18:55:58Z)
• ST-WebAgentBench: A Benchmark for Evaluating Safety and Trustworthiness in Web Agents [3.09793323158304]
  Existing benchmarks measure only whether an agent finishes a task, ignoring whether it does so safely or in a way enterprises can trust.<n>We introduce textbftextscST-WebAgentBench, a suite for evaluating web agent ST across realistic enterprise scenarios.<n>Each of its 222 tasks is paired with ST policies, concise rules that encode constraints, and is scored along six dimensions (e.g., user consent, robustness)
  arXiv  Detail & Related papers  (2024-10-09T09:13:38Z)
• Dissecting Adversarial Robustness of Multimodal LM Agents [70.2077308846307]
  We manually create 200 targeted adversarial tasks and evaluation scripts in a realistic threat model on top of VisualWebArena.<n>We find that we can successfully break latest agents that use black-box frontier LMs, including those that perform reflection and tree search.<n>We also use ARE to rigorously evaluate how the robustness changes as new components are added.
  arXiv  Detail & Related papers  (2024-06-18T17:32:48Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.