Tool use / function calling

AgentAuditor

AgentAuditor: Human-Level Safety and Security Evaluation for LLM Agents

Superseded baseline#29 of 55 most-superseded · first seen May 31, 2025

Cited as a baseline — critiqued by newer work, not yet beaten on a benchmark here

1 papers critique it · 0 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites AgentAuditor as a baseline.

However, autonomous agents require dynamic safety monitoring over each tool invocation steps to enable timely intervention against emerging risks.
ToolSafe: Enhancing Tool Invocation Safety of LLM-based agents via Proactive Step-level Guardrail and Feedback

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.