Translation in progress Some entries still include text in the original language.
Oct 12 · Monday
All papers Found 16 papers Another 254 papers do not have category tags yet and are not included in filtered results.
“
About this collection Collection in progress We are building a large AI safety paper collection to help researchers find work related to their interests. Papers are collected automatically andfiltered only for relevance to AI safety , without judging quality, novelty or reading value. The default “Annotated first” shows papers with tags or published paper analyses first. Each group is ordered by original publication date, newest first, with undated papers last. “Newest first” uses publication dates only. This order is not a quality ranking.
One-line summaries, tags and paper analyses are prepared by AI from the papers. Language models may have systematic biases and produce hallucinations.We plan to carry out human annotation and review to reduce errors in this content.
Proposes digital lock compilation, encrypting executable networks into host weights to hide arbitrary capabilities
Published Oct 10
Evaluated models Gemma 3 270M Abstract The authors use a digital lock to hide an executable capability and state that elicitation on the chemistry and parity tasks did not recover the capability.The authors construct a class of backdoors that encrypt an executable network into the weights, used to discuss the limits of white-box elicitation audits under supply chain attacks.
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 Oct 5
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。
提出 UnAct，仅以前向激活干预从分类器中遗忘指定类别
Published Oct 3
Evaluated models ResNet-18 , ViT-B/16 Abstract UnAct 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。
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 Sep 30
Evaluated models Stable Diffusion v1.5 , Stable Diffusion XL , Stable Diffusion v3.5 and others4 Abstract NDDL 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。
提出 NEEDLE，用权重正交化在已知触发器时去除大模型后门
Published Sep 30
Evaluated models Gemma-3-4B-IT , Qwen3-4B-Instruct-2507 , Gemma-3-12B-IT Abstract NEEDLE 用闭式权重编辑去掉已知触发器的后门，并保住拒绝子空间。
用 SAE 定位后门触发机制并分离检测与因果控制
Published Sep 8
Evaluated models Gaperon-1B , Gaperon-8B , Gaperon-24B Abstract Gaperon 语言切换后门可分成检测、残差路由和后期语言跟踪三类 SAE 特征，只有部分能改行为。作者用稀疏自编码器分解 Gaperon 中无害的语言切换后门，区分识别触发的特征与真正能抑制或诱发该行为的特征。固定触发使英语提示在 Gaperon-1B 与 8B 上续写成法语或德语。
解释后验安全的权重几何脆弱性，并提出预训练持续安全共训练
Published Sep 7
Evaluated models Qwen-2.5-7B , Qwen-2.5-7B-Instruct , Llama-3-8B and others15 Abstract 论文将后验安全脆弱性归因于正交抑制，并提出通过预训练持续共训练实现具备微调抗性的安全防护。
提出 MROP，在发射端推理时纯化输入以抵御训练投毒后门
Published Sep 1
Evaluated models DeepJSCC , SwinJSCC Abstract MROP 不重训练，在发射端纯化补丁后门。MROP 是 deep JSCC 无线图像传输上的推理时输入纯化，对象是矩形补丁后门。
提出 AMS，用激活分离与方向相似度检测安全训练改动
Published Aug 6
Evaluated models Llama-3.1-8B-Instruct , Llama-3.2-3B-Instruct , Llama-3.2-1B-Instruct and others14 Abstract AMS 按激活签名把安全改动分四类，前三类可由两级扫描覆盖，行为微调会漏。AMS 是面向开放权重模型、部署前使用的激活扫描器，用来看安全训练被拿掉或改写之后，激活空间里还留着什么几何痕迹。
用通用扰动对恶意软件分类器及漂移检测器做规避与后门投毒
Published Aug 4
Evaluated models MLP 100-100 , CADE (CAE 512-384-256-128) , CADE (CAE 512-128-32-8) and others6 构造可通过计费审计的 token 虚报，使服务商系统性超收
Published May 28
Evaluated models PALACE domain-adapted auditor Abstract 不诚实提供方可在三种 token 审计上虚报并通过检查，作者认为需改用提供方不控制的证据。不透明 LLM 的按 token 计费里，现有审计核验的是提供方可控的证据，不诚实提供方可以虚报并通过检查。 。
复评前沿模型代码生成中的包名幻觉与抢注攻击面
Published May 17
Evaluated models Claude Sonnet 4.6 , Claude Haiku 4.5 , GPT-5.4-mini and others5 Abstract 评估显示前沿模型包幻觉率收窄但威胁仍在，作者识别出跨模型通用攻击面并呼吁联合防御。
提出 SeedHijack，劫持采样 PRNG 以强制选定 token
Published May 8
Evaluated models GPT-2 (124M) , Qwen2-1.5B-Instruct , DeepSeek-R1-1.5B and others5 Abstract 采样层 PRNG 被替换以强制选 token。SeedHijack 把 LLM 采样用的 PRNG 输出换成攻击者选定的数，从而指定下一个 token。权重、输入和 logits 都不改。防御是用硬件 QRNG 替换该随机源。
Comparing jailbreak fine-tuning and weight orthogonalization unalignment on harmful assistance and usefulness
Published Apr 3
Evaluated models Qwen3-4B-Instruct-2507 , Llama-3.1-8B-Instruct , Qwen2.5-14B and others6 用 Lee-Valiant 稳健均值缓解 steering 数据集污染
Published Mar 4
Evaluated models Llama-3.2-3B-Instruct , Mistral-7b-Instruct-v0.3 , OLMo-2-1124-7B-Instruct Abstract 数据污染会改 steering，协调注入最强。作者研究 contrastive steering 的训练数据被污染后，推理时的行为控制还剩多少，以及稳健均值能否接住 。
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 Feb 3
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。
You have reached the end