{"slug": "runtime-verification-of-multiple-natural-language-criteria-for-agent-governance", "title": "Runtime Verification of Multiple Natural Language Criteria for Agent Governance", "summary": "A new arXiv paper (2611.23074v1) introduces the VFM, a tree-attention verifier that scores N natural language criteria against an agent's output in parallel and returns independent per-criterion ternary verdicts of satisfied, violated, or inapplicable. The verifier 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 and beats DynaGuard-4B on trace-level failure detection in the DynaBench multi-criterion benchmark. On a single H100, VFM-4B evaluates 1,000 criteria against a 4K-token context in under 2 seconds at under 11 GB of VRAM, which the authors present as making runtime agent governance practical.", "body_md": "arXiv:2611.23074v1 Announce Type: new \nAbstract: 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.", "url": "https://wpnews.pro/news/runtime-verification-of-multiple-natural-language-criteria-for-agent-governance", "canonical_source": "https://www.machinebrief.com/news/runtime-verification-of-multiple-natural-language-criteria-f-1ht3", "published_at": "2026-11-19 05:00:00+00:00", "updated_at": "2026-11-19 05:43:04.932436+00:00", "lang": "en", "topics": ["ai-agents", "ai-safety", "ai-research", "large-language-models", "ai-policy"], "entities": ["VFM", "VFM-4B", "CriteriaBank", "DynaBench", "DynaGuard-4B", "GPT-5.4", "H100", "arXiv"], "also_reported_by": [], "takeaway": "VFM-4B checks 1,000 natural language criteria against a 4K-token context in under 2 seconds on one H100 using under 11 GB of VRAM, beating GPT-5.4 on a synthetic governance benchmark and DynaGuard-4B on DynaBench trace-level failure detection.", "recommendation": {"advice": "For runtime agent governance, pretrain a tree-attention verifier on criterion-verdict tuples like CriteriaBank's 394K examples and score criteria in parallel; VFM-4B handled 1,000 criteria per 4K-token context in under 2s on one H100.", "playbook": "evals-testing", "playbook_url": "https://wpnews.pro/playbooks/evals-testing"}, "alternates": {"html": "https://wpnews.pro/news/runtime-verification-of-multiple-natural-language-criteria-for-agent-governance", "markdown": "https://wpnews.pro/news/runtime-verification-of-multiple-natural-language-criteria-for-agent-governance.md", "text": "https://wpnews.pro/news/runtime-verification-of-multiple-natural-language-criteria-for-agent-governance.txt", "jsonld": "https://wpnews.pro/news/runtime-verification-of-multiple-natural-language-criteria-for-agent-governance.jsonld"}}