跳到正文
arXiv cs.AI· Silviu Pitis, Parand A. Alamdari, Jessica Tang, Toryn Q. Klassen, Sheila A. McIlraith·· 2 小时前AI 评分51

VFM 论文提出面向智能体治理的多自然语言标准运行时验证方法

Runtime Verification of Multiple Natural Language Criteria for Agent Governance

AI 导读

研究者提出树注意力验证器 VFM,可并行对 N 条自然语言治理标准输出独立的三元判定(满足、违反或不适用),模型在 CriteriaBank 预训练,该语料含 394K 条元组、覆盖 321K 条不同标准。

正文 · 原文

View PDF HTML (experimental)

Abstract:AI agents are governed by hundreds of natural language criteria, yet existing verifiers cannot efficiently evaluate an agent's output against a large set of such criteria and report which are satisfied, violated, or inapplicable. We introduce the VFM, a tree-attention verifier that encodes the shared context and target text, and scores N natural language criteria in parallel, returning independent per-criterion ternary verdicts. The VFM is pretrained on CriteriaBank, a corpus of 394K (context, target, criterion, verdict) tuples spanning over 321K distinct criterion strings. A finetuned VFM-4B outperforms GPT-5.4 on a synthetic governance benchmark. On the DynaBench multi-criterion benchmark, it outperforms DynaGuard-4B on trace-level failure detection, and its per-criterion verdicts localize the violated rule and support focused escalation to a larger chain-of-thought judge. We show that CriteriaBank pretraining improves cross-domain transfer, particularly in the low-data regime, while matching a strong direct-finetuning control at full data. On a single H100, VFM-4B evaluates 1,000 criteria against a 4K-token context in under 2s at <11 GB VRAM, making it suitable for runtime verification.
Comments: Accepted at NeurIPS 2026
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as: arXiv:2611.23074 [cs.CL]
  (or arXiv:2611.23074v1 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2611.23074

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Silviu Pitis [view email]
[v1] Tue, 17 Nov 2026 17:01:00 UTC (1,989 KB)

来源:arXiv cs.AI · arxiv.org