Skip to content

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

Oct 13 · Tuesday

All papers

Found 96 papers
Filter5

Attack type

Jailbreaks5Prompt injection19Poisoning and backdoors49Model tampering16Extraction and theft10Detection evasion3

Attack technique

Gradient and representation optimization5Automated search4Encoding and obfuscation3Defense-targeted7Dormant triggers16

Defense type

Model hardening21Inference-time intervention3Detection and filtering10Instruction–data separation4Permissions and approval4

Impact

Harmful content22Unauthorized actions9Data and privacy leaks12Output manipulation41Availability and reliability6Oversight failure3

System type

Models and APIs59LLM applications3AI Agent35coding agent4Multi-agent systems27Embodied systems and robotics3

Attacker access

Black-box3White-box7

Attack surface / entry point

User input223Web pages and interfaces69Documents and messages88Tool responses106Tool and skill descriptions63Code repositories and dependencies29Retrieval corpus38Memory writes42Inter-agent messages28Images, audio and video51Training data68Weights and gradients71Distribution channels63Inference services and hardware8

Layer

Prompt layer4Model layer9Application layer31Training and data layer56

System component

MCP and skills6Memory4Guardrails and judges7Reasoning chains3Permissions and sandboxing8Fine-tuning and open weights35Vision6Multimodal10
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 surface / entry point：Inter-agent messagesAttack surface / entry point：Training dataLayer：Model layerLayer：Application layerLayer：Training and data layerClear all

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

  1. Analysis and theoryarXiv 2610.13051

    Space beats statistics, retraining beats ranking: auditing training labels under label-only backdoors

    Auditing training labels for label-only backdoors, comparing spaces and statistics, and judging by retraining

    arXiv
    2610.13051
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    ResNet-32
    AbstractLabel inspection must be done in a space that has not been fitted to these labels, and judged by retrained models rather than detection scores.

    The authors ask: can a defender without trusted data decide which training labels to keep, and how much clean data this destroys, and what the final model accuracy is.

    Paper analysisRead analysis
  2. AttacksarXiv 2610.12644

    Cancerous backdoor attack: progressive activation via invisible tokens and a stage controller.

    Proposes the Cancerous backdoor, which progressively corrupts reasoning using invisible stage tokens.

    arXiv
    2610.12644
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Llama-3.1-8B-Instruct, Qwen2.5-7B-Instruct, Mistral-7B-Instruct-v0.3
    AbstractHidden tokens plus a stage controller make the backdoor progressively corrupt reasoning upon triggering, and the authors believe that looking only at the answer will miss it.

    Cancerous Backdoor Attack changes LLM backdoors from one-time activation to multi-stage reasoning corruption that accumulates progressively with triggers.

    Paper analysisRead analysis
  3. DefensesarXiv 2610.10345

    SLDR：通过选择性层恢复与动态路由防御恶意微调

    提出 SLDR，用选择性层恢复与动态路由防御恶意微调

    arXiv
    2610.10345
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    Llama3.1-8B-Instruct, Llama3-8B-Instruct, Qwen2.5-7B-Instruct and others4
    Abstract作者称 SLDR 能压低有害输出并保持下游效用。

    SLDR 是针对恶意微调的微调后防御：在已对齐模型上做有符号层探测，只在敏感分数最大与最小的两层训练 LoRA，再用 lmax 的最后 token 表示决定是否启用适配器。

    Paper analysisRead analysis
  5. DefensesarXiv 2610.09941

    OrthoPurify 用一步正交投影净化被后门污染的视觉语言模型

    提出 OrthoPurify，用正交投影清除视觉语言模型适配器中的劫持方向

    arXiv
    2610.09941
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    LLaVA-1.5-7B, Qwen3-VL-8B, LLaVA-1.5-13B and others5
    Abstract一次投影去掉劫持方向。

    OrthoPurify 是部署前的 LVLM 后门净化方法，用一次正交投影去掉被劫持的权重更新方向。

    Paper analysisRead analysis
  6. AttacksarXiv 2610.07723

    Researchers propose answer-side backdoors: letting the model self-generate trigger words to bypass safety refusals.

    提出回答端后门，让模型自生成触发词以在干净有害请求下绕过拒答

    arXiv
    2610.07723
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Llama-2-7B-chat-hf, Mistral-7B-Instruct-v0.3, Gemma-7B-it and others4

    Abstract论文提出了多轮对话中回答端自生成触发词绕过安全拒绝的后门攻击，指出输入端防护存在盲区。

    Paper analysisRead analysis
  7. AttacksarXiv 2610.07654

    研究揭示策略蒸馏的安全后门风险

    测量带后门教师经 OPD 向干净学生迁移潜伏后门

    arXiv
    2610.07654
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Qwen2.5-1.5B-Instruct, Qwen2.5-3B-Instruct, Qwen2.5-7B-Instruct
    Abstract作者称 OPD 可在压低无触发词有害率时迁入后门。

    带后门的外部教师经 OPD，可在压低无触发词有害率的同时，把触发条件下的有害行为传给初始干净学生。。

    Paper analysisRead analysis
  8. InterpretabilityarXiv 2610.07518

    TRACE：仅凭 checkpoint 更新审计大语言模型有害目标

    提出 TRACE，仅凭检查点更新审计有害合规微调目标

    arXiv
    2610.07518
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3, Qwen3-8B and others6
    Abstract作者称有害合规 SFT 在检查点更新中留下可排序坐标，并据此做只读权重审计。

    作者在检查点更新里寻找有害合规 SFT 的目标组成，并把读数做成不查询模型、不看未知训练数据的审计分 TRACE。

    Paper analysisRead analysis
  9. DefensesarXiv 2610.06636

    差分隐私混合公共数据集提升隐私学习效果

    提出 DP-MixMin，私有学习公开数据混合并做差分隐私微调

    arXiv
    2610.06636
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    ResNet-18, GPT-2 Medium
    AbstractDP-MixMin 私下学公开混合再 DP 微调，胸片 macro AUC 与邮件困惑度优于均匀混合。

    DP-MixMin 是一条端到端的差分隐私训练流程：先私下决定多个公开数据集的预训练混合，再预训练并在敏感数据上 DP 微调。作者称这是首个这样的流程。

    Paper analysisRead analysis
  10. AttacksarXiv 2610.05793

    研究者训练出按多智能体拓扑触发的代码后门模型

    提出拓扑条件后门，使代码模型在推断多智能体部署时植入漏洞

    arXiv
    2610.05793
    Published
    Layer
    Training and data layer
    Setting
    coding agent
    Evaluated models
    Qwen2.5-7B-Instruct, Qwen2.5-Coder-7B-Instruct

    Abstract论文证实了拓扑条件后门的可训练性与泛化性，展示了常规评测盲区，并提出差异化拓扑审计作为检测防御方向。

    Paper analysisRead analysis
  11. DefensesarXiv 2610.05640

    Research reveals that the CaMeL defense fails in multi-agent systems and proposes multi-CaMeL

    提出 multi-CaMeL，在智能体调用边界隔离可信指令与非可信数据

    arXiv
    2610.05640
    Published
    Layer
    Application layer
    Setting
    Multi-agent systems
    Evaluated models
    Claude Fable 5, GPT-5.5, GPT-4.1 and others6

    Abstract论文针对多智能体中单体防御不组合的边界清洗问题，提出 multi-CaMeL 协议彻底阻断了 MultiAgentDojo 上的注入攻击。

    Paper analysisRead analysis
  12. DefensesarXiv 2610.04426

    UnAct：无需梯度的定向激活干预实现类别遗忘

    提出 UnAct，仅以前向激活干预从分类器中遗忘指定类别

    arXiv
    2610.04426
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    ResNet-18, ViT-B/16
    AbstractUnAct 用前向激活做类别遗忘。

    UnAct 是一种事后类别遗忘：删除请求可以只有少量该类图像，且没有标签和保留数据。它不用梯度，只对遗忘图像做前向。编辑对象是分类器所读的后期单元。

    Paper analysisRead analysis
  13. Analysis and theoryarXiv 2610.02886

    研究揭示无触发投毒审计缺口：事实答对不等于决策正确

    揭示无触发虚假训练下事实回答与下游决策的审计差距

    arXiv
    2610.02886
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Qwen2.5-0.5B, Qwen2.5-1.5B, Qwen2.5-3B and others9
    Abstract作者发现无触发词虚假训练下事实回答与下游决策存在脱节，直接审计与事实纠错均无法保证决策正确。

    论文是一项针对语言模型在无触发词虚假信息训练下行为表现的实证分析研究。

    Paper analysisRead analysis
  14. DefensesarXiv 2610.02349

    MIRROR：面向多智能体通信的法定人数完整性防御

    提出 MIRROR，用多路径摘要多数表决保障多智能体通信完整性

    arXiv
    2610.02349
    Published
    Layer
    Application layer
    Setting
    Multi-agent systems
    Evaluated models
    Gemma-4-31B-it, Gemini 2.5 Pro
    AbstractMIRROR 靠多数路径。

    MIRROR 是 LLM 多智能体通信层上的多路径法定人数完整性原语，只处理不透明字节，不解释消息内容。要处理的是 Agent-in-the-Middle：中介合法终结 TLS 后可改在途消息，而语义校验有推理开销和误拦，签名又要求私钥管理。做法是生成结束后复制同一规范化载荷。

    Paper analysisRead analysis
  15. DefensesarXiv 2610.02005

    贝叶斯辩证论证（BDA）：面向持续对抗下多 LLM 委员会的可校准聚合

    提出 BDA，按辩证动作可靠性加权聚合多 LLM 议会以抵抗持续对抗

    arXiv
    2610.02005
    Published
    Layer
    Application layer
    Setting
    Multi-agent systems
    AbstractBDA 用零额外调用的可靠性加权，给出可校准后验，并在持续对抗下反转不可靠席位。

    BDA 是一种不另调 LLM 的多 LLM 议会聚合方法，同时处理置信度校准和持续不可靠席位。

    Paper analysisRead analysis
  16. Evaluation and benchmarksarXiv 2610.03153

    EvoRiskBench: An evolutionary benchmark for runtime security risks of workspace agents

    构建 EvoRiskBench 评测工作区 Agent 的运行时安全风险

    arXiv
    2610.03153
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    GPT-5.6 Sol, DeepSeek-V4-Pro-0813, Claude Opus 5

    Abstract论文构建了工作空间智能体运行时风险基准 EvoRiskBench，发现高执行力智能体在 IPI 下面临较高安全风险。

    Paper analysisRead analysis
  18. AttacksarXiv 2610.01367

    研究：高质量数据筛选挡不住投毒

    提出 Bi-QSTO，优化能穿过质量筛选并在微调中削弱安全对齐的投毒样本

    arXiv
    2610.01367
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Llama2-7B-Chat, Llama3-8B-Instruct, Qwen2-7B-Instruct and others5
    Abstract质量筛选挡不住部分高分样本的安全削弱。

    Bi-QSTO 研究一个实际训练缺口：基于质量的数据选择会先于微调决定样本去留，而以往微调投毒多假设毒样本直接进入训练集。已对齐 LLM 的拒答会被少量有害或看似无害的微调样本削弱。作者用三个问题串起实验：选择能否排除毒样本。

    Paper analysisRead analysis
  19. AttacksarXiv 2610.01365

    ReGap：无需真实隐私数据即可通过微调重新提取模型隐私关联

    提出 ReGap，用自生成伪监督微调恢复模型中的隐私关联

    arXiv
    2610.01365
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Attacker
    White-box, Gray-box
    Evaluated models
    GPT-2 Small, GPT-2 Medium, GPT-2 Large and others6

    Abstract提出无真实私有监督的恢复攻击 ReGap，实验显示微调可通过自生成数据唤醒模型潜在隐私关联，突显后适应隐私风险。

    Paper analysisRead analysis
Load more