收录解读
多轮分解攻击将有害目标分散在多个看似合理的请求和工具调用中,使传统单步检测失效,对工具型LLM代理构成严重安全威胁。
本文提出ReDiR,一种生成时防御方法,通过轨迹级安全证据条件化动作生成:将当前轨迹压缩为潜在安全表示并注入冻结的基础模型,利用同模型、跨视图监督学习,无需单独的安全模块即可在生成过程中整合跨轮次安全信息。
在两个代理安全基准、三个模型家族和八个保留工具域上的实验表明,ReDiR将攻击成功率降至8%以下,并可迁移至未见过的工具域,同时保持良性保真度且计算开销低。
该方法为多轮代理安全提供了一种新的思路,但缺少与强基线(如在线推理防御)的详细性能对比,且实验细节(如基准名称、具体开销数值)未在摘要中充分披露,限制了对可重复性和鲁棒性的全面评估。
原始摘要
Tool-using LLM agents extend security risks beyond generated text to actions that affect external systems. Under multi-turn decomposition attacks, a harmful objective can be distributed across individually plausible requests and tool calls, becoming apparent only from the accumulated trajectory. Existing defenses either rely on auxiliary online reasoning to recover long-horizon security evidence or assess actions after generation, often incurring additional inference cost or depending on runtime-specific action representations. We propose \emph{Reassembling Distributed Risk} (ReDiR), a generation-time defense that conditions action generation on trajectory-level security evidence. Before each action, ReDiR compresses the current trajectory into a compact latent safety representation and injects it into the frozen base model. The representation is learned through same-model, cross-view supervision, where safe behavior from an explicit task view provides supervision for recovering distributed safety evidence from the original multi-turn trajectory. This design enables ReDiR to integrate cross-turn security information directly within the generation process without relying on a separate action-level safety module. We evaluate ReDiR on two agent-safety benchmarks across three model families and eight held-out tool domains. ReDiR reduces attack success rates to below 8\%, transfers to unseen tool domains, and preserves benign fidelity with low computational overhead.