Skip to content

Translation in progress Some entries still include text in the original language.

Oct 13 · Tuesday

All papers

Found 22 papers
Filter6

Attack type

Jailbreaks4Prompt injection77Poisoning and backdoors22Detection evasion4Malicious use3

Attack technique

Dormant triggers8

Defense type

Instruction–data separation4

Impact

Unauthorized actions8Output manipulation10

System type

LLM applications3AI Agent19coding agent8

Attack surface / entry point

User input24Web pages and interfaces12Documents and messages17Tool responses16Tool and skill descriptions35Code repositories and dependencies7Retrieval corpus25Memory writes35Inter-agent messages4Training data2Weights and gradients4Distribution channels37

Layer

Model layer9Application layer22Training and data layer46

System component

MCP and skills5Memory10RAG4Fine-tuning and open weights3
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 227papers have no tags yet and are not included in filtered results.
Attack type：Poisoning and backdoorsAttack surface / entry point：Web pages and interfacesAttack surface / entry point：Code repositories and dependenciesAttack surface / entry point：Training dataAttack surface / entry point：Weights and gradientsLayer：Application layerClear all

Another 227papers do not have category tags yet and are not included in filtered results.

  1. AttacksarXiv 2610.11932

    A Fudan team measures the citation fragility of AI search: ordinary posts can enter citations and answer text

    Measures citation fragility in AI search, demonstrating that ordinary public posts can enter citations and answers.

    arXiv
    2610.11932
    Published
    Layer
    Application layer
    Setting
    LLM applications
    Evaluated models
    Grok, Doubao, ChatGPT and others10

    AbstractThe authors claim the paper quantifies vulnerabilities in AI search caused by citation preferences and low posting thresholds; tests show public web posts can penetrate AI citations, and they call for auditing the citation supply chain.

    Paper analysisRead analysis
  2. AttacksarXiv 2610.09264

    研究者提出 PackHallu：经规则文件提示注入对编码 Agent 发起包幻觉攻击

    提出 PackHallu，经规则文件提示注入诱导编码 Agent 换用攻击者指定包

    arXiv
    2610.09264
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    Qwen2.5-Coder-7B, Qwen3-Coder-30B, Devstral-Small-2-24B and others19
    AbstractPackHallu 诱导规则文件换包。

    PackHallu 研究一种针对编程智能体规则文件的提示注入：让智能体在本应使用合法依赖时，改导入攻击者指定并控制的包。

    Paper analysisRead analysis
  3. AttacksarXiv 2610.07510

    PersistBD: attackers can strengthen backdoors before release so that they maintain high attack success rates even after developer SFT and RL.

    提出 PERSISTBD，发布前强化供应链后门使其经良性 SFT 与 RL 后仍外发凭据

    arXiv
    2610.07510
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    Qwen2.5-Coder-3B-Instruct, Qwen2.5-Coder-7B-Instruct, Qwen3-Coder-30B-A3B-Instruct
    Abstract总结了论文关于后门在 SFT/RL 中存活的发现、两因子机制、PERSISTBD 方法与核心数字。

    本文研究第三方 LLM 智能体在开发者进行良性监督微调（SFT）和强化学习（RL）后训练时的后门持久性风险。

    Paper analysisRead analysis
  5. AttacksarXiv 2609.39788

    研究：多智能体系统潜在通信链路可被训练用于提升有害顺从

    提出只训练或篡改多智能体潜在通信链路来提高有害遵从

    arXiv
    2609.39788
    Published
    Layer
    Application layer
    Setting
    Multi-agent systems
    Evaluated models
    Llama-3.2-3B-Instruct, Qwen2.5-3B-Instruct, Gemma-3-1B-IT and others7
    Abstract潜在链路本身即可改变冻结智能体系统的安全，奖励既能攻击也能修复。

    学习得到的潜在通信链路会改变由冻结、对齐智能体组成的系统是否遵从有害请求。。作者研究三种拓扑：双智能体、三级顺序协作，以及双专家加汇总。

    Paper analysisRead analysis
  6. AttacksarXiv 2609.37468

    论文提出针对 Agentic RAG 的检索器后门攻击与 DOU 隐藏方法

    提出检索器后门攻击操纵智能体搜索，并用 DOU 逃避检测

    arXiv
    2609.37468
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    E5, Contriever-MSMARCO

    Abstract论文展示了通过恶意检索器操纵智能体多轮搜索并利用诱饵重塑表征隐蔽后门的可行性与系统风险。

    Paper analysisRead analysis
  7. Evaluation and benchmarksarXiv 2609.32616

    Cave-Bench：虚假指控下 LLM 智能体最高 60.06% 的运行会破坏正确工作

    提出 CAVE-BENCH，评测智能体在虚假指责下破坏已完成的正确工作

    arXiv
    2609.32616
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    Claude-Sonnet-5, Claude-Opus-5, GPT-5.6-Sol and others14

    Abstract论文提出了评估智能体在虚假指责下破坏已有正确成果的 CAVE-BENCH，发现强模型易推翻已知证据，提出了门控缓解方案。

    Paper analysisRead analysis
  8. Evaluation and benchmarksarXiv 2609.28915

    研究提出 ACE 语料库，评估内核级证据对 Agent 安全检测的有效性

    构建 ACE 语料，评测内核级证据对 Agent 攻击检测的有效性

    arXiv
    2609.28915
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    AdaBoost, XGBoost, TabPFN and others10
    AbstractACE 用配对的内核与应用层证据表明，syscall 痕迹可判别智能体攻击，两层组合通常更好。

    ACE 是一份把智能体会话的内核 syscall 痕迹与应用层清单、转录配对起来的检测语料和基准。。

    Paper analysisRead analysis
  10. AttacksarXiv 2609.17817

    论文：用投毒基准污染自改进编码 Agent，令其后续版本写出漏洞代码

    提出基准投毒攻击，使自修改编码 Agent 跨代写出含漏洞代码

    arXiv
    2609.17817
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    Claude Sonnet 4.5, Qwen3.5-397B, gpt-oss-120b and others5

    Abstract论文报告了恶意基准能污染自修改代码智能体的进化过程并持续输出漏洞，警示系统设计需纳入安全约束。

    Paper analysisRead analysis
  11. AttacksarXiv 2609.02774

    CodePoisonRAG：针对检索增强代码生成的知识投毒攻击

    提出 CodePoisonRAG，对检索增强代码生成的知识库投毒以植入漏洞

    arXiv
    2609.02774
    Published
    Layer
    Application layer
    Setting
    LLM applications
    Evaluated models
    Qwen 3.5 9B, Code Llama 13B, DeepSeek-Coder-V2 16B

    AbstractCodePoisonRAG 展示了 RACG 外部知识库的新攻击面，在较低投毒率下可定向诱发指定漏洞并部分抵御防御。

    Paper analysisRead analysis
  14. AttacksarXiv 2608.06862

    SynChain：诱导计算机使用 Agent 自建攻击链并跨任务持久化

    提出 SYNCHAIN，诱导计算机使用 Agent 自建跨任务持久攻击链

    arXiv
    2608.06862
    Published
    Layer
    Application layer
    Setting
    Browser and computer use, coding agent
    Evaluated models
    Qwen3.5-9B, Llama-3.1-8B, Ministral-3-8B

    Abstract论文揭示了 CUA 自合成工件带来的内部供应链威胁，验证了潜伏载荷的跨任务传播能力与防御局限。

    Paper analysisRead analysis
  15. Evaluation and benchmarksarXiv 2607.14611

    论文评测 Claude Code 与 Codex 记忆文件中的提示注入风险

    评测编码 Agent 记忆文件中提示注入的跨会话影响

    arXiv
    2607.14611
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    Claude Haiku 4.5, Claude Opus 4.7, GPT-5.2 and others4
    Abstract已植入记忆文件的载荷可以跨会话影响 Claude Code 与 Codex，攻击成功与载荷是否留下是两种不同的失败。

    Bad Memory 在沙箱合成工作区里评测文件型持久记忆上的提示注入，比较载荷能否改变当前会话，以及能否在后续会话中留下或叠加。

    Paper analysisRead analysis
  17. AttacksarXiv 2606.04329

    研究者系统研究 LLM 智能体记忆投毒攻击并提出 MPBench 基准

    系统研究 LLM 智能体记忆投毒并构建基准 MPBench

    arXiv
    2606.04329
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    Meta-Llama-3.1-70B-Instruct, Deberta-v3-base, Llama-3.1-8B-Instruct and others5

    Abstract系统研究了智能体记忆投毒风险，揭示了写入渠道漏洞，构建了 MPBench 基准并指出提示注入防御的局限。

    Paper analysisRead analysis
  18. AttacksarXiv 2605.15338

    Hidden in Memory: Latent Memory Poisoning of LLM Agents

    提出潜伏记忆投毒，诱导 Agent 把外部文档写成持久伪造记忆

    arXiv
    2605.15338
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    GPT-5.4, GPT-5.5, Claude Sonnet 4.6 and others12
    Abstract持久记忆成为跨会话攻击面：通用载荷可写入伪造记忆，并在相关后续任务中影响行为。

    sleeper memory poisoning 研究带持久记忆的 LLM 助手：外部文档能否让助手存下一则伪造的用户记忆，并在原文档消失后的新会话里被取回、用来改变回复或动作。

    Paper analysisRead analysis
  19. AttacksarXiv 2604.02623

    论文提出 eTAMP 攻击：网页内容注入可跨会话污染 Web Agent 记忆

    提出 eTAMP，用网页内容跨会话投毒 Web Agent 轨迹记忆并诱导越权操作

    arXiv
    2604.02623
    Published
    Layer
    Application layer
    Setting
    web agent
    Evaluated models
    GPT-5-mini, GPT-5.2, GPT-OSS-120B and others6
    Abstract论文提出了基于环境注入的跨任务记忆投毒攻击 eTAMP，作者报告智能体在环境压力下易感性增加且能力与安全并不协同。

    该论文评估了基于大语言模型的 Web 智能体在利用历史交互轨迹进行记忆增强时的安全脆弱性。

    Paper analysisRead analysis
  20. DefensesarXiv 2603.12644

    Revealing security threats and defense architectures for autonomous agents: an OpenClaw case study

    Reviews threats to autonomous agents and proposes a full-lifecycle defense blueprint, FASA

    arXiv
    2603.12644
    Published
    Layer
    Application layer
    Setting
    AI Agent
    AbstractUsing OpenClaw as an example, it divides risks into three layers and uses FASA to describe defense from input to system execution.

    This is a threat review and defense architecture report using OpenClaw as a case study, not a scored attack or defense evaluation.

    Paper analysisRead analysis
Load more