Sep 2026· Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing· Vol abs/2609.28197· 0 citations· 52 references
Computer Science
TL;DR
Decoupled Proactive Safety Monitoring is formalized along three dimensions: whether to intervene, when to intervene, and what the risk is, to address proactive intervention limitations in Large Language Models.
Abstract
As Large Language Models (LLMs) evolve into autonomous agents that alter real-world states, ensuring operational safety across multi-step workflows has become a critical challenge. While recent work has moved beyond single-turn evaluation toward multi-turn paradigms, key limitations persist: step-level methods treat actions in isolation, missing how risks accumulate, while trajectory-level evaluations operate post-hoc, offering no opportunity for timely intervention. To address these limitations, we formalize Decoupled Proactive Safety Monitoring along three dimensions: whether to intervene, when to intervene, and what the risk is. We introduce PASTABench, a benchmark of 1,139 multi-turn trajectories spanning 5 risk categories and 13 subcategories. We further propose the Optimal Intervention Window (OIW), anchored by annotated Earliest-Signal and Trigger turns, to quantify intervention timeliness. Evaluation of 16 LLMs reveals that proactive intervention remains largely unsolved, with the best model achieving only 40.74% optimal-timing interventions. Fine-grained diagnosis further uncovers pervasive lexical overfitting: competitive safety scores of smaller models mask keyword hypersensitivity rather than genuine risk comprehension, as their proactive capability largely collapses once hazard vocabulary is neutralized.
The transition from Large Language Models (LLMs) to agents shifts safety stakes from toxic text to irreversible environmental harm. While current defenses remain largely retrospective, proactive runtime intervention is bottlenecked by the lack of large-scale, causally-consistent data. We propose PROACT-Agent, a framewo...
Ding Jia, Wei Liu, Xiang-Long Du et al.· 0 citations
Large language model agents can invoke tools fluently, but enterprise workflows demand more than selecting the right tools: actions must strictly comply with organizational policies, tool feedback often conceals hidden side effects under partial observability, and long-horizon tasks require persistent state tracking ac...
AURA-Eval, a framework combining controlled augmentation with granular diagnosis of behavior in tool-use trajectories, is introduced, showing that LLM agents engage in unsafe behavior more often when no safe fulfillment path exists.
Ruoxi Shang, Christina-Maria Androna, Orfeas Menis-Mastromichalakis et al.· 0 citations
DynSTEER bridges the granularity gap through stage-wise dynamic evaluation that segments rollouts at key execution nodes and adapts its multi-level review strategy based on stage-level results; it compiles a path-tolerant milestone graph from available task inputs to preserve diverse legal paths without reference leaka...
Zhichao Shi, Xuhui Jiang, Wen-Jie Zhang et al.· 1 citation
This work introduces AGNI, an automated pipeline that extracts trajectory-relevant assumptions, injects targeted environmental changes, and validates that the resulting novel tasks remain solvable and highlights a gap between task competence and adaptive capability and motivate environmental variation as a core dimensi...
This work investigates how open-source guard models represent two types of trajectory-level agentic harms and finds that they are linearly readable inside the model, even though guard models predict no better than chance on pairs that differ only in the called tool's schema.