跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2610.09142 · v3

Finding Side-Effect Loopholes in AppWorld and WorkArena Verifiers

首次提交 2026-10-07 05:34
版本更新 2026-11-16 08:53
雷达观测 2026-11-17 17:12
arXiv ↗PDF ↗

摘要

Language-model agents act for people inside software, sending payments or editing records. Benchmarks such as AppWorld and WorkArena grade them with automated verifiers, whose verdicts also become reinforcement learning (RL) rewards, so a lenient verifier rewards mistakes. Verifier audits mostly hunt for runs that pass without doing the task; we ask about runs that do it and also change something nobody asked for. We mutation-test both verifiers with hand-written edits to correct runs, designed from the verifier source; an end-state check confirms each false positive. All confirmed false positives were unrequested changes; planted omissions and wrong values produced none. AppWorld's tests passed a duplicate write in all six tasks of two templates, and three RL methods that train on one of these templates would fully reward it; one count assertion per template rejected all six duplicates while passing every correct run. WorkArena's form verifier, called once after submission, passed all 36 edited runs that filled an unrequested field. Checked after every field, it caught extra fields filled first or in drop-downs but still passed all 8 runs that filled a text or reference field last. Both verifiers check that the requested change happened but not reliably that nothing else changed.

作者备注:24 pages. Camera-ready version, accepted as a poster at the NeurIPS 2026 workshop "Who Verifies the Agents? Toward Reliable Agent Development". OpenReview: https://openreview.net/forum?id=uR6EWsGE48. Code and evidence: https://github.com/OpenAdaptAI/side-effect-loopholes

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

本站首次收录为 v3,暂无更早版本可对照。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v3 · 2026-11-16 08:53 · 最新观测版本

Finding Side-Effect Loopholes in AppWorld and WorkArena Verifiers

Language-model agents act for people inside software, sending payments or editing records. Benchmarks such as AppWorld and WorkArena grade them with automated verifiers, whose verdicts also become reinforcement learning (RL) rewards, so a lenient verifier rewards mistakes. Verifier audits mostly hunt for runs that pass without doing the task; we ask about runs that do it and also change something nobody asked for. We mutation-test both verifiers with hand-written edits to correct runs, designed from the verifier source; an end-state check confirms each false positive. All confirmed false positives were unrequested changes; planted omissions and wrong values produced none. AppWorld's tests passed a duplicate write in all six tasks of two templates, and three RL methods that train on one of these templates would fully reward it; one count assertion per template rejected all six duplicates while passing every correct run. WorkArena's form verifier, called once after submission, passed all 36 edited runs that filled an unrequested field. Checked after every field, it caught extra fields filled first or in drop-downs but still passed all 8 runs that filled a text or reference field last. Both verifiers check that the requested change happened but not reliably that nothing else changed.

查看此版本解读 →查看此版本原文 ↗