收录解读
LLM智能体在长时序工作流中执行shell命令、修改代码、安装依赖并与远程服务交互,越来越多地将外部内容中的对抗指令当作合法任务指令,面临间接提示注入(IPI)风险。现有评估多集中于孤立恶意字符串、单步工具调用或窄配置场景,缺乏对任务生命周期内多攻击面的统一评测。因此,构建贴近真实代理工作流的IPI基准成为智能体安全的重要问题。
INFUSE提出了一个端到端基准,包含40个真实任务和56个攻击场景,按五个威胁类别组织,并将攻击注入到任务执行过程中的现实交互面,而不仅限于初始提示或单独工具调用。该设计支持对长流程代理在代码修改、依赖安装、远程服务交互等环节暴露的IPI风险进行系统评测,并附带开源代码。
作者评估了来自GPT、Claude、Gemini、Llama、Gemma和Qwen系列的19个LLM,生成41,824条对抗轨迹(其中26,220条无防御)和720条良性参考运行。LLM评测的平均攻击成功率为19.6%,启发式验证为15.4%,各模型ASR-L在3.2%至41.9%之间;现有guard模型保护有限,提示级指令加固将平均攻击成功率从58.2%降至14.3%,但存在模型相关的安全-效用权衡。
未达paradigm:全量主要使用Mini-SWE-Agent及单判官,单判官与人类一致性κ仅0.53;跨harness迁移条件化于既往成功攻击,58.2%降至14.3%只适用于受控高风险子集,不能建立普适防御规律。
原始摘要
LLM-based agents increasingly execute long-horizon, end-to-end workflows involving shell commands, code modification, dependency installation, and interaction with remote services. This growing autonomy expands their exposure to indirect prompt injection (IPI), where adversarial instructions embedded in external content are interpreted as legitimate task directives. Existing evaluations, however, often focus on isolated malicious strings, single-step tool calls, or narrowly scoped agent configurations. We introduce INFUSE, an end-to-end benchmark for evaluating IPI throughout the lifecycle of long-horizon agent workflows. INFUSE comprises 40 realistic tasks and a structured taxonomy of 56 attack scenarios spanning five threat categories. Attacks are injected through realistic interaction surfaces encountered during task execution, extending evaluation beyond attacks placed only in initial prompts or isolated tool interactions. We evaluate 19 LLMs from the GPT, Claude, Gemini, Llama, Gemma, and Qwen families, producing 41,824 adversarial trajectories (26,220 without defenses) and 720 benign reference runs. Across the benchmark, mean attack success reaches 19.6% under LLM-based evaluation and 15.4% under heuristic verification, with per-model ASR-L ranging from 3.2% to 41.9%. Existing guard models provide limited protection, while prompt-level instructional hardening reduces mean attack success from 58.2% to 14.3%, revealing substantial but model-dependent security-utility trade-offs. Our code is available at https://github.com/judydnguyen/infuse.