Skip to content

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

Oct 12 · Monday

All papers

Found 17 papers
Filter3

Attack type

Jailbreaks193Prompt injection17Poisoning and backdoors11Model tampering19Extraction and theft36Detection evasion47Denial of service3Malicious use11

Attack technique

Gradient and representation optimization3Cross-modal payloads4

Impact

Output manipulation10

System type

Models and APIs13AI Agent3

Attack surface / entry point

User input11Images, audio and video5

Layer

Prompt layer12Model layer5Application layer224

System component

Guardrails and judges3Vision4Multimodal5
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 265papers have no tags yet and are not included in filtered results.
Attack type：Prompt injectionLayer：Prompt layerLayer：Model layerClear all

Another 265papers do not have category tags yet and are not included in filtered results.

  1. InterpretabilityarXiv 2610.12361

    Study: legal reasoning models citing legal provisions does not equal relying on legal provisions.

    Uses premise permutation and hidden-layer tracing to audit whether legal chain-of-thought truly relies on the cited legal provisions.

    arXiv
    2610.12361
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Qwen3-8B, Qwen3-14B, Llama-3.1-8B and others7
    AbstractUsing counterfactual substitution of authorities and internal commitment tracing, the authors find that legal CoT rulings depend on authorities only to a limited extent and are vulnerable to prompt injection manipulation.

    This study conducts an empirical audit of the causal truthfulness of legal citations in chain-of-thought reasoning of large language models in the legal domain.

    Paper analysisRead analysis
  2. DefensesarXiv 2610.08678

    研究揭示投机解码的安全代价并提出 SecureSD 加固方法

    测量有损投机解码的安全代价，并提出 SecureSD 收紧早期验证

    arXiv
    2610.08678
    Published
    Layer
    Model layer
    Setting
    AI Agent
    Attacker
    Black-box
    Evaluated models
    Qwen3-8B, Qwen3-32B, Llama-3.1-8B-Instruct and others6
    Abstract有损推测解码的早期偏离更快抬高越狱与注入风险。

    作者研究有损推测解码在换取速度时，是否会比效用更快地损害越狱与提示注入防护，并给出按位置校正的 SecureSD。

    Paper analysisRead analysis
  4. DefensesarXiv 2610.02413

    研究：分类后缀可提升激活探针的分布外恶意输入检测

    用分类后缀提升激活探针对分布外恶意输入的检测

    arXiv
    2610.02413
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Llama-3.1-8B, Qwen3.5-9B, Gemma-4-12B
    Abstract作者称排序增益来自分类格式，低 FPR 精度来自点名准则。

    用户轮后的分类后缀，被作者用作激活探针的分布外检测干预：排序看格式，低 FPR 精度看是否点名恶意/良性。

    Paper analysisRead analysis
  5. InterpretabilityarXiv 2609.37737

    研究用因果激活修补定位模型服从提示注入指令的决策层

    用因果激活修补定位模型服从提示注入的决策层

    arXiv
    2609.37737
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Qwen3-4B, Qwen3-14B, Qwen3-32B and others7
    Abstract合规的因果杠杆集中在网络后三分之一的低秩子空间，该层也是混淆下检测最好的位置。

    这篇工作用逐层激活修补定位提示注入合规发生在网络何处，并把该位置用于检测验证。。

    Paper analysisRead analysis
  7. Evaluation and benchmarksarXiv 2609.33401

    论文评估 System One 模型在 Agent 安全决策中的可靠性、校准与选择性自动化

    评测 System One 模型对 Agent 安全输入的分类可靠性与校准

    arXiv
    2609.33401
    Published
    Layer
    Prompt layer
    Setting
    LLM applications
    Evaluated models
    Jev 1.13, Laya, English checkpoint, Decider, approximately 2B and others4
    Abstract离线评测表明，安全判定的总体分数和校准不能代替分组误差和带预算的自动决策检查。

    Liu 离线评测 System One 模型能否为智能体安全输入给出可自动化的概率决策。问题是分类是否可靠，以及概率能否在指定漏检和误拦上限内支持放行、拦截或复核。

    Paper analysisRead analysis
  8. Evaluation and benchmarksarXiv 2609.29429

    Jev 被用于零样本检测十类对齐失败，中位 AUROC 达 0.886

    提出 RLCDAlignBench，零样本检测多种对齐失败

    arXiv
    2609.29429
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    Jev (jev-1.13.0)
    Abstract一次通用问题已能给多数对齐失败排序，大的误差来自状态和标签。

    RLCDAlignBench 测量用强化学习做校准决策的 Jev，能否当作对齐失败的零样本检测器。

    Paper analysisRead analysis
  9. Evaluation and benchmarksarXiv 2609.04859

    MM-IFEval-Pro：面向视觉语言模型的多语言抗攻击指令遵循基准

    提出抗攻击多语言视觉指令遵循基准 MM-IFEval-Pro

    arXiv
    2609.04859
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    Qwen3-VL-8B-Instruct, InternVL3.5-8B, MiniCPM-V4.5 and others5
    AbstractMM-IFEval-Pro 以规则验证评测中英指令遵循与视觉劫持，GRPO 提高两基座平均分。

    MM-IFEval-Pro 是面向 VLM 的中英多模态指令遵循基准，并配有可用规则计算奖励的 GRPO 训练集。

    Paper analysisRead analysis
  10. DefensesarXiv 2609.01046

    HiveTraceGuard-Pro：面向提示注入、越狱与对抗混淆的紧凑生成式护栏

    提出生成式护栏 HiveTraceGuard-Pro，检测提示注入、越狱与混淆

    arXiv
    2609.01046
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    HiveTraceGuard-Pro (0.6B)
    Abstract0.6B 俄英护栏在自建俄语干净集和 PI-RU 上得分最高，聚合 key 居第三，过阻断和回复精度仍是作者标出的短板。

    HiveTraceGuard-Pro 是面向请求与回复审核的 0.6B 生成式护栏，由 Qwen3-0.6B 经 LoRA 微调，在俄语和英语上训练，只输出一个 safe/unsafe 裁决。

    Paper analysisRead analysis
  11. AttacksarXiv 2609.01023

    Akrasia：针对推理型代码 LLM 的隐蔽后门攻击

    提出隐蔽后门攻击 AKRASIA，在推理代码模型提示中植入触发器并伪装思维链

    arXiv
    2609.01023
    Published
    Layer
    Prompt layer
    Setting
    coding agent
    Evaluated models
    Claude-Sonnet-5, GPT-5.5, Gemini-3.5-flash and others7

    Abstract论文针对推理代码大模型提出了隐蔽推断期后门攻击 AKRASIA，揭示了模型思维链可被利用进行欺骗性伪装的安全隐患。

    Paper analysisRead analysis
  12. AttacksarXiv 2606.06244

    研究者提出 Ghostwriter 攻击：用伪造证据注入操纵 LLM 观点

    提出 Ghostwriter 攻击，用伪权威证据条件注入操纵模型观点

    arXiv
    2606.06244
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    GPT-4o, Claude-3.7-Sonnet, DeepSeek-V3 and others10
    Abstract作者称攻击未被消除。

    Ghostwriter 是一种针对上下文窗口的观点引导攻击：把误导陈述改写成伪权威文本，再在相关查询中让目标模型把该观点写进回答。

    Paper analysisRead analysis
  14. AttacksarXiv 2604.14604

    AudioHijack 框架对 13 个语音大模型实现隐蔽音频提示注入

    提出 AudioHijack，用不可感知对抗音频对语音模型做上下文无关提示注入

    arXiv
    2604.14604
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    SpeechGPT, GLM-4-Voice, VITA-Audio and others15

    Abstract提出了 AudioHijack 框架，揭示了 LALM 对听觉提示注入的脆弱性与传统防御的局限。

    Paper analysisRead analysis
  16. AttacksarXiv 2602.15927

    研究者提出面向多轮对话的视觉记忆注入攻击 VMI

    提出 VMI，用扰动图像在多轮对话中潜伏并仅在触发话题输出预定消息

    arXiv
    2602.15927
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    Qwen2.5-VL-7B-Instruct, Qwen3-VL-8B-Instruct, LLaVA-OneVision-1.5-8B-Instruct and others5
    AbstractVMI 让扰动图像在多轮 LVLM 中潜伏，作者称触发话题出现后仍可输出预定消息。

    VMI 是面向多轮 LVLM 的定向图像攻击：第三方发布带小扰动的图像，良性用户把它当作普通输入。

    Paper analysisRead analysis
You have reached the end