A Trajectory-Based Safety Audit of Clawdbot (OpenClaw)
Research paper by Tianyu Chen and collaborators. Published 2026-02-16. Topics: tool-using personal AI agent, trajectory-centric evaluation, agent-safety benchmarks, ATBench, LPS-Bench.
Resource information
- Authors
- Tianyu Chen, Dongrui Liu, Xia Hu, Jingyi Yu, Wenjie Wang
- Research topics
- tool-using personal AI agent, trajectory-centric evaluation, agent-safety benchmarks, ATBench, LPS-Bench, automated trajectory judge, AgentDoG-Qwen3-4B, interaction trajectories, safety assessment
- Publication date
- 2026-02-16
- Publication
- arXiv
- Last record update
- Nov 3, 2026
