用隐式推理提升智能体安全评估，解决长轨迹中风险证据稀疏问题。
DRAFT: Task Decoupled Latent Reasoning for Agent Safety
- 将安全判断拆分为提取器与推理器两阶段，隐式聚合证据。
- 在ASSEBench和R-Judge上准确率从63.27%提升至91.18%。
- 适合研究智能体安全、长序列监控的开发者使用。
工具型大模型智能体的兴起使安全监控从输出审核转向对长而嘈杂的交互轨迹进行审计，其中风险关键证据稀疏，传统二值监督难以实现责任分配。为此，我们提出DRAFT（Task Decoupled Latent Reasoning for Agent Safety），一种将安全判断解耦为可训练两阶段的隐式推理框架：提取器将完整轨迹压缩为紧凑连续的隐式草稿，推理器联合关注草稿与原始轨迹以预测安全性。DRAFT通过在隐空间中执行证据聚合，避免了损失严重的显式摘要-判断流水线，支持端到端可微训练。在ASSEBench和R-Judge等基准上，DRAFT持续优于强基线，平均准确率从63.27%（LoRA）提升至91.18%，且学习到更具可分性的表征。消融实验表明提取器与推理器之间存在显著协同效应。总体而言，DRAFT表明，在读出前进行连续隐式推理是长上下文、稀疏证据条件下实现鲁棒智能体安全的有效路径。
原文摘要 · Abstract (English)
The advent of tool-using LLM agents shifts safety monitoring from output moderation to auditing long, noisy interaction trajectories, where risk-critical evidence is sparse-making standard binary supervision poorly suited for credit assignment. To address this, we propose DRAFT (Task Decoupled Latent Reasoning for Agent Safety), a latent reasoning framework that decouples safety judgment into two trainable stages: an Extractor that distills the full trajectory into a compact continuous latent draft, and a Reasoner that jointly attends to the draft and the original trajectory to predict safety. DRAFT avoids lossy explicit summarize-then-judge pipelines by performing evidence aggregation in latent space, enabling end-to-end differentiable training.Across benchmarks including ASSEBench and R-Judge, DRAFT consistently outperforms strong baselines, improving accuracy from 63.27% (LoRA) to 91.18% averaged over benchmarks, and learns more separable representations. Ablations demonstrate a clear synergy between the Extractor and the Reasoner.Overall, DRAFT suggests that continuous latent reasoning prior to readout is a practical path to robust agent safety under long-context supervision with sparse evidence.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。