When Tools Lie: Reliability of Mathematical Agents Under Corrupted Tool Feedback

Key points

• Yet tools can fail silently, returning plausible but incorrect results.1
• How well can agents detect and correct corrupted tool call outputs?1
• We study this through a controlled corruption framework where a hidden interceptor replaces tool call results with plausible incorrect information on targeted problems.1
• We evaluate agents across 31 problems under four verification designs including no verification (baseline), mandatory same-context reflection, optional fresh-context verification, and optional structural verification.1

Sources (1)

• [1]When Tools Lie: Reliability of Mathematical Agents Under Corrupted Tool Feedback
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 10:28 AM

  Mathematical problem solving often requires deterministic computational steps that agents delegate to tools and implicitly trust.

  Yet tools can fail silently, returning plausible but incorrect results.

Extractive summary: sentences quoted from the sources.