Skip to content

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

Oct 13 · Tuesday

All papers

Found 14 papers

Another 201papers do not have category tags yet and are not included in filtered results.

  1. 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

    Published
    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
  2. AttacksarXiv 2610.06049

    研究者提出仅修改 SAE 解码器的后门攻击,可在不改动语言模型的情况下植入行为

    提出只改 SAE 解码器的后门,在冻结语言模型上植入代码插入

    Published
    Evaluated models
    Gemma 3 4B Instruct, Qwen3 1.7B, Llama 3.1 8B Instruct
    Abstract只改解码器的 SAE 可带上代码插入后门,且若干常规质量指标变化不大。

    只改 SAE 解码器、冻结编码器与语言模型,即可把代码插入行为放进推理时的辅助组件。作者把这视为 SAE 供应链问题:checkpoint 一旦替换某一层激活,就参与后续计算,而不只是解释工具。

    Paper analysisRead analysis
  3. AttacksarXiv 2610.05894

    研究者提出闭环偏见注入攻击

    提出闭环激活引导,在扩散语言模型去噪过程中注入定向偏见

    Published
    Attacker
    Gray-box
    Evaluated models
    LLaDA-8B-Instruct, Dream-v0-Instruct-7B, LLaDA-MoE-7B-A1B-Instruct

    Abstract通过去噪中间概率反馈动态调节残差流转向,证明扩散语言模型在推理服务栈中存在闭环偏见注入脆弱性。

    Paper analysisRead analysis
  4. 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

    Published
    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
  5. DefensesarXiv 2609.39548

    NDDL: Backdoor Defense for Text-to-Image Models with Normal Diffusion Dynamics

    提出 NDDL,用良性扩散轨迹检测文生图后门提示并定位触发 token

    Published
    Evaluated models
    Stable Diffusion v1.5, Stable Diffusion XL, Stable Diffusion v3.5 and others4
    AbstractNDDL 从良性扩散轨迹学习正常转移,用动态不一致检测后门并定位触发 token。

    NDDL 是面向文生图扩散模型的后门防御:防御者只有白盒模型访问和良性提示词,没有后门先验,需要区分后门提示词并找出触发 token。

    Paper analysisRead analysis
  6. DefensesarXiv 2609.38909

    Purdue 提出 Pact:将欺骗作为行为遗忘

    提出 PACT,遗忘推理模型在压力下的欺骗行为并保留上下文理解

    Published
    Attacker
    White-box
    Evaluated models
    DeepSeek-R1-Distill-Qwen-32B, QwQ-32B

    Abstract论文将大模型欺骗形式化为上下文条件行为,提出 PACT 框架在消除欺骗的同时保留了上下文遵循能力。

    Paper analysisRead analysis
  7. DefensesarXiv 2608.02271

    Z-PEFT: Zero-shot backdoor detection for parameter-efficient fine-tuning based on canonical spectral features

    Propose Z-PEFT, which detects PEFT backdoors in a zero-shot manner using per-head spectral features.

    Published
    Evaluated models
    Z-PEFT, Llama-2-7B, Llama-2-13B and others6
    Paper analysisRead analysis
  8. AttacksarXiv 2602.17837

    TFL: Targeted Bit-Flip Attack against Large Language Models

    Proposes TFL targeted bit flipping so that selected prompts output false answers while unrelated queries remain correct

    Published
    Evaluated models
    Qwen3-8B, DeepSeek-R1-Distill-Qwen-14B, LLaMA-3.1-8B-INSTRUCT
    AbstractTFL searches weight bits on three LLMs to make specified prompts produce target keywords while constraining degradation on unrelated tasks.

    TFL is a framework for targeted bit flips on LLM main memory weights: the attacker wants a selected prompt to output a specified false answer keyword while limiting degradation on unrelated queries.

    Paper analysisRead analysis
  9. DefensesarXiv 2602.03085

    Triggers in the Haystack: Extracting and Reconstructing LLM Backdoor Triggers

    提出推理扫描器,在仅有模型文件时检测后门并重构触发词

    Published
    Evaluated models
    gemma-3-270m-it, DeepSeek-R1-Distill-Qwen-1.5B, Phi-4-mini-instruct and others8
    Abstract用记忆泄漏和内部信号重构触发词。

    这是一个只做推理的 sleeper agent 扫描器:在不知触发词和目标行为时,检测因果语言模型里的条件后门并重构触发词。。问题在于,后门只在触发词出现时偏离基线,而开放仓库和网页投毒使检测方通常没有触发词、目标或激活语料。

    Paper analysisRead analysis
  10. AttacksarXiv 2602.01658

    Efficient Adversarial Attacks on High-Dimensional Offline Bandits

    提出三类攻击,训练前扰动奖励模型以劫持离线 bandit

    Published
    Evaluated models
    Image Reward, Aesthetic Model
    Abstract公开奖励权重可被小扰动劫持离线 bandit,打乱日志只部分缓解。

    作者研究离线 bandit 在奖励模型被事先改权重时,还会不会选出最优臂。

    Paper analysisRead analysis
You have reached the end