arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.28013cs.CRcs.AIcs.LG
收藏

Kill-Chain Canaries: 阶段级跟踪跨攻击表面和模型安全层级的提示注入

Kill-Chain Canaries: Stage-Level Tracking of Prompt Injection Across Attack Surfaces and Five Production LLMs

Massachusetts Institute of Technology(麻省理工学院);University of Chicago(芝加哥大学)

Haochuan Kevin Wang, Zechen Zhang

AI总结:

研究通过跟踪加密令牌四个阶段,揭示提示注入是管道-架构问题,发现写节点位置是最高安全决策,所有防御因通道不匹配失效,且隐形白字体PDF负载可匹配或超越可见文本ASR。

摘要:

多代理LLM系统正在进入生产阶段,处理文档、管理流程、代表用户行动,但其对提示注入的韧性仍仅用二元评估:攻击是否成功?这使架构师缺乏硬化真实管道所需诊断信息。我们引入了kill-chain canary方法,通过四个阶段(EXPOSED -> PERSISTED -> RELAYED -> EXECUTED)在950次运行、五种前沿LLM、六个攻击表面和五个防御条件下追踪加密令牌。结果将提示注入重新定义为管道-架构问题:每个模型完全暴露,但下游结果各异——Claude在内存写入阶段阻止所有注入(0/164 ASR),GPT-4o-mini传播率达53%,DeepSeek在相同模型的不同表面表现0%/100%。三个发现对部署至关重要:(1)写节点位置是最高安全决策——通过验证模型路由写入可消除传播;(2)所有四个防御因通道不匹配在至少一个表面失效,无需对抗性适应;(3)隐形白字体PDF负载可匹配或超过可见文本ASR,意味着渲染层筛查不足。这些动态直接适用于生产:机构投资者和金融公司已运行NLP管道处理财报、SEC文件和分析师报告——文档摄入流程正迁移到LLM代理。代码、运行日志和工具已公开发布。

英文摘要:

Multi-agent LLM systems now read documents, web pages and tool results on behalf of users, yet their resistance to prompt injection is usually reported as one number: did the attack succeed? We introduce a kill-chain canary method that plants a unique token in every injected payload and records the furthest of four stages it reaches (Exposed -> Persisted -> Relayed -> Executed), across 950 runs, five production LLMs, six attack surfaces, and five defense conditions. Exposure was 100% among runs that called the tool; the outcomes differ downstream. Claude Haiku 4.5 and Claude Sonnet 4.5 executed none of their 164 text-surface attacks, and in the text relay the canary token never appeared in a memory write (0/40); GPT-4o-mini executed 53% of its attacks. Four findings follow. (1) A Claude writer kept the canary token out of shared memory in every relay run we report; one cross-model pairing (Claude writer, GPT-4o-mini reader, n = 3) is consistent with this protecting the reader, and other pairings were not tested. (2) As readers, the Claude models executed 0/40 raw pre-seeded injections, but Claude Haiku 4.5 executed 2/3 injections relayed by GPT-4o-mini; whether relayed injections are harder to refuse than raw ones is an open question. (3) DeepSeek Chat went from 0/24 on pre-seeded memory to 8/8 on tool results, scenarios that also differ in task and payload format; white-text PDF payloads, invisible on the rendered page, succeeded at least as often as visible ones. (4) pi_detector and write_filter failed on channels they do not inspect, spotlighting failed on content it wraps, and write_filter blocked the PDF relay but not the text relay, a difference we cannot explain. Code and run logs are publicly released: https://github.com/KevinChunye/prompt_injection

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑