arXiv cs.AI· Jonghyun Lee, Moonsu Kim, Sunghwan Park·· 12 小时前AI 评分31
智能体的自验证器何时可信？在搜索压力与 Oracle 访问下压力测试反事实理由验证
When Can You Trust an Agent's Own Verifier? Stress-Testing Counterfactual Rationale Verification Under Search Pressure and Oracle Access
AI 导读
研究提出用压力比 ρ 和四个可证伪探针来压力测试智能体的自验证器，并在 LLM 智能体设计机器学习流水线的场景中检验了自研的 Rationale Consistency Score（RCS）。在先前配置下验证器对搜索目标影响极小（ρ≪1），部分因准确率门控，其鲁棒性此前未被真正测试。当智能体可在提交前探测验证器时，它会保留能通过检查的声明，而留出集忠实度下降，这一下降在循环内指标中不可见。
正文 · 原文
Abstract:Automated agent design increasingly trusts a single score to decide whether a change is an improvement -- but a score optimized aggressively enough ceases to measure what it was meant to. When an agent also explains its design, that explanation is a second signal we can check without human labels: negate each stated claim, re-run, and observe whether performance drops. We ask whether such a verifier stays reliable once an agent optimizes against it. Our contribution is not a new verifier but a way to stress-test one: a pressure ratio $\rho$ measuring how much the verifier steers the agent's search, and four falsifiable probes covering four failure modes -- giving no signal, being too weak to matter, being gamed by stating fewer claims, and being queried before the agent commits. We apply the test to our own verifier, the Rationale Consistency Score (RCS), inside an LLM agent that designs machine-learning pipelines, with one generator model, six tasks, and three seeds. First, in our prior configuration the verifier barely affects the search objective ($\rho \ll 1$), partly because of an accuracy gate, so its robustness had not been tested; it still shapes the reported artifact through a final-selection tie-break that $\rho$ does not register. Second, a capability control shows the agent shortens its claim lists when instructed, yet search does not, even at $\rho$ over 100 times the prior level; since feedback omits claim volume and horizons are short, this null result is largely structural rather than evidence of robustness. Third, when the agent can probe the verifier before committing, it keeps claims that pass while held-out faithfulness falls -- a decline invisible to in-loop metrics, though partly confounded by how RCS scores small pipelines. We offer these as preliminary evidence and propose a reporting standard: audit a verifier with $\rho$ and these probes before trusting it.
| Comments: | 17 pages, 4 figures, 8 tables. Accepted at the NeurIPS 2026 Workshop "Who Verifies the Agents? Toward Reliable Agent Development" |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.22779 [cs.AI] |
| (or arXiv:2610.22779v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.22779 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: JongHyun Lee [view email]
[v1]
Fri, 16 Oct 2026 00:41:30 UTC (626 KB)
来源：arXiv cs.AI · arxiv.org