Tagged: llm-as-judge

LLM-Judge Reliability for Agent Safety Violations on Attacked Trajectories

A research project that audits the auditors: measuring how position, verbosity, and self-preference biases distort LLM-judge judgments of safety violations...