WeClawArena: An Auditable Sandbox and Benchmark for Cross-User Agents Collaboration and Security in Human-Centered Agent Networks
The paper introduces a benchmark that tests whether personal agents can collaborate across user-owned workspaces without leaking data or accepting bad authority.
WeClawArena includes 124 base tasks across six cross-user domains, expanded into 620 variants. Each task has a benign control plus four attack-vector versions. The sandbox logs peer messages, tool calls, resource operations, governed decisions, and final workspace states for audit. It reports utility and attack success separately, with checks for privacy leakage, poisoned evidence, invalid authority paths, and task breakdown. HF Daily Papers' note
WeClawArena includes 124 base tasks across six cross-user domains, expanded into 620 variants. Each task has a benign control plus four attack-vector versions. The sandbox logs peer messages, tool calls, resource operations, governed decisions, and final workspace states for audit. It reports utility and attack success separately, with checks for privacy leakage, poisoned evidence, invalid authority paths, and task breakdown. HF Daily Papers' note
score 5