AgentProv: Auditing Agentic LLM API Providers via Tool-use Policy Probes

AgentProv compares the tool-selection behavior of a suspect endpoint with a trusted instance of the claimed model. Visit the project page for the method, results, and paper.

Recommended citation: Wang, X., Zhao, B., Backes, M., Boenisch, F., & Dziedzic, A. (2026). "AgentProv: Auditing Agentic LLM API Providers via Tool-use Policy Probes." EMNLP 2026.
Download Paper