# ActBench: Self-Evolving Benchmark of Behavioral Safety in Cowork Agents

arXiv: https://arxiv.org/abs/2608.09476 · HTML: https://mlsanity.com/paper/2608.09476/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use.
- Builds on 18 of its 29 references that are in the map.
- First posted on arXiv 10 Aug 2026.

## Placed in

- LLM agents and tool use / [Agent threats, guardrails and governance](https://mlsanity.com/agents/agent-threats-guardrails-and-governance/)
- Synthetic data and self-improvement for LLMs / [Safety of self-evolving systems](https://mlsanity.com/synth/safety-of-self-evolving-systems/)

## Builds on (most cited first, 18 of 18)

- [AgentPoison: Red-teaming LLM Agents via Poisoning Memory or Knowledge Bases](https://mlsanity.com/paper/2407.12784/) (2024)
- [SecAlign: Defending Against Prompt Injection with Preference Optimization](https://mlsanity.com/paper/2410.05451/) (2024)
- [WASP: Benchmarking Web Agent Security Against Prompt Injection Attacks](https://mlsanity.com/paper/2504.18575/) (2025)
- [EIA: Environmental Injection Attack on Generalist Web Agents for Privacy Leakage](https://mlsanity.com/paper/2409.11295/) (2024)
- [Memory Injection Attacks on LLM Agents via Query-Only Interaction](https://mlsanity.com/paper/2503.03704/) (2025)
- [DataSentinel: A Game-Theoretic Detection of Prompt Injection Attacks](https://mlsanity.com/paper/2504.11358/) (2025)
- [OS-Harm: A Benchmark for Measuring Safety of Computer Use Agents](https://mlsanity.com/paper/2506.14866/) (2025)
- [RedCode: Risky Code Execution and Generation Benchmark for Code Agents](https://mlsanity.com/paper/2411.07781/) (2024)
- [Can LLMs Separate Instructions From Data? And What Do We Even Mean By That?](https://mlsanity.com/paper/2403.06833/) (2024)
- [MCPTox: A Benchmark for Tool Poisoning Attack on Real-World MCP Servers](https://mlsanity.com/paper/2508.14925/) (2025)
- [MobileSafetyBench: Evaluating Safety of Autonomous Agents in Mobile Device Control](https://mlsanity.com/paper/2410.17520/) (2024)
- [AgentDoG: A Diagnostic Guardrail Framework for AI Agent Safety and Security](https://mlsanity.com/paper/2601.18491/) (2026)
- [AttriGuard: Defeating Indirect Prompt Injection in LLM Agents via Causal Attribution of Tool Invocations](https://mlsanity.com/paper/2603.10749/) (2026)
- ["Humans welcome to observe": A First Look at the Agent Social Network Moltbook](https://mlsanity.com/paper/2602.10127/) (2026)
- [HarmfulSkillBench: How Do Harmful Skills Weaponize Your Agents?](https://mlsanity.com/paper/2604.15415/) (2026)
- [Quantifying Conversation Drift in MCP via Latent Polytope](https://mlsanity.com/paper/2508.06418/) (2025)
- [Red-Teaming Agent Execution Contexts: Open-World Security Evaluation on OpenClaw](https://mlsanity.com/paper/2605.11047/) (2026)
- [Securing the AI Agent: A Unified Framework for Multi-Layer Agent Red Teaming](https://mlsanity.com/paper/2606.31227/) (2026)

## Abstract (arXiv)

Cowork agents may complete benign tasks while disclosing protected data, manipulating unauthorized state, invocate unauthorized API. We define behavioral safety and introduce ActBench, a self-evolving benchmark that evaluates such behavior risk from execution trajectories rather than final responses. Each case pairs a benign task with an adversarial variant that preserves its instruction, configuration, initial state, rating model, and trusted records while injecting a task-reachable payload. ActBench contains 600 cases from 213 scenarios, spanning 15 risk behaviors, six execution spaces, and 48 web-service APIs.To move beyond static payloads, we propose a reward-guided beam search method that jointly optimizes attack effectiveness and task utility, while reflection diagnoses failed execution checkpoint and guides payload revision. Besides, we propose a dual evidence verification mechanism that verifies agent execution safety and utility through log evidence and LLM-based trajectory evidence.We evaluate 15 LLMs and 6 open-source cowork agents over 24,000 trajectories. Under a fixed harness, attack success rates ranges from 10.1% to 94.4% across models, while under a fixed base model, they range from 73.7% to 94.4% across agents.These results show greater variation across models than agent harness, while attacks remain highly successful across all tested harnesses.Our benchmark is released at: https://github.com/zjuicsr/ActBench.