PROACT-Agent: Progressive Runtime Oversight and Active Circuit-breaking for Real-Time Safety

2026-09-28

AI Summary

This study addresses the bottleneck in real-time proactive safety defense for LLM agents, which stems from the scarcity of causally consistent data during environment interactions. To this end, we propose a high-fidelity trajectory synthesis framework. Methodologically, we design a progressive trajectory unfolding mechanism coupled with reasoning-enhanced causal correction to mitigate safety drift. Furthermore, multi-model arbitration annotation, bilingual benchmark construction, and cultural localization strategies are introduced to enforce causal consistency constraints. Experimental results demonstrate that the proposed approach achieves an F1 score of 91.46% on unsafe instances and reduces the attack success rate from 20.82% to 0.40%, thereby establishing efficient real-time safety guardrails for LLM agents.

Abstract

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 framework for synthesizing high-fidelity trajectories to enable real-time guardrails. We identify a critical "safety drift" in prior benchmarks, where lenient annotation paradigms fail to enforce temporal consistency. PROACT-Agent addresses this through: (1) Progressive Trajectory Unrolling to reveal risks hidden in long-context interactions; (2) Reasoning-Augmented Causal Rectification to enforce monotonic causal consistency; and (3) Culturally-Aware Data Localization for cross-border robustness. We introduce PROACT-Bench, a bilingual safety benchmark with 155,780 states labeled through multi-model adjudication. Evaluating updated context before the next LLM inference, the trained guard achieves 91.46% unsafe-class F1 and 90.63% exact-boundary detection under complete source holdout. In AgentDojo, it reduces non-DoS targeted attack success from 20.82% to 0.40%.

Problem

• LLM Agents Safety
• Runtime Intervention
• Safety Drift
• Causal Consistency
• Real-time Guardrails

Innovation

• Progressive Trajectory Unrolling
• Causal Rectification
• Real-time Guardrails
• Safety Drift
• PROACT-Bench