Helpful Agent Meets Deceptive Judge: Understanding Vulnerabilities in Agentic Workflows
This addresses critical reliability issues in feedback-based AI systems for developers and researchers, though it is incremental in analyzing known vulnerabilities.
The paper systematically analyzes vulnerabilities in agentic workflows where LLM judges provide deceptive feedback, revealing that even strong agents often switch correct answers after misleading critiques. It introduces a framework and benchmark (WAFER-QA) to evaluate robustness, showing distinct behavioral patterns in models over multiple interaction rounds.
Agentic workflows -- where multiple large language model (LLM) instances interact to solve tasks -- are increasingly built on feedback mechanisms, where one model evaluates and critiques another. Despite the promise of feedback-driven improvement, the stability of agentic workflows rests on the reliability of the judge. However, judges may hallucinate information, exhibit bias, or act adversarially -- introducing critical vulnerabilities into the workflow. In this work, we present a systematic analysis of agentic workflows under deceptive or misleading feedback. We introduce a two-dimensional framework for analyzing judge behavior, along axes of intent (from constructive to malicious) and knowledge (from parametric-only to retrieval-augmented systems). Using this taxonomy, we construct a suite of judge behaviors and develop WAFER-QA, a new benchmark with critiques grounded in retrieved web evidence to evaluate robustness of agentic workflows against factually supported adversarial feedback. We reveal that even strongest agents are vulnerable to persuasive yet flawed critiques -- often switching correct answers after a single round of misleading feedback. Taking a step further, we study how model predictions evolve over multiple rounds of interaction, revealing distinct behavioral patterns between reasoning and non-reasoning models. Our findings highlight fundamental vulnerabilities in feedback-based workflows and offer guidance for building more robust agentic systems.