arXiv cs.AI· Kavienan Jegatheesan, Gayathri Lihinikaduarachchi·· 10 天前AI 评分30
当工具说谎：工具反馈被污染下数学智能体的可靠性
When Tools Lie: Reliability of Mathematical Agents Under Corrupted Tool Feedback
AI 导读
研究用隐藏拦截器把数学智能体的工具调用结果替换为看似合理的错误信息，在 31 道题上测试四种验证设计。无验证时准确率从 100% 跌至 72.4%，强制同上下文反思可完全恢复到 100%，可选验证只有在模型主动调用时才提升准确率。显式检测后完全重启问题在 100% 的情况下恢复成功，表明验证器可用性与验证策略是数学智能体可靠性的两个独立组成部分。
正文 · 原文
Abstract: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. How well can agents detect and correct corrupted tool call outputs? We study this through a controlled corruption framework where a hidden interceptor replaces tool call results with plausible incorrect information on targeted problems. 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. Without verification, corruption causes dramatic accuracy loss, from 100% down to 72.4%. Mandatory reflection fully recovers this performance to 100%. Optional verification improves accuracy only when models actively invoke it. Our results show that checking frequency is strongly associated with robustness differences, while unequal invocation prevents a controlled comparison of verifier quality. A supporting recovery experiment shows that full problem restart succeeds in 100% of cases after explicit detection. These findings demonstrate that verifier availability and verification policy are separate components of mathematical-agent reliability. Mandatory policies enforce verification while optional policies depend on the model's own choice to invoke it.
| Comments: | 8 pages, 2 figures, 3 tables. Accepted to the 6th Workshop on Mathematical Reasoning and AI (MathAI) at NeurIPS 2026 |
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Software Engineering (cs.SE) |
| Cite as: | arXiv:2610.08097 [cs.CR] |
| (or arXiv:2610.08097v1 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2610.08097 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Kavienan Jegatheesan [view email]
[v1]
Tue, 6 Oct 2026 10:28:19 UTC (1,115 KB)
来源：arXiv cs.AI · arxiv.org