Translation in progress Some entries still include text in the original language.
Oct 12 · Monday
All papers Found 92 papers Another 240 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.
Auditing training labels for label-only backdoors, comparing spaces and statistics, and judging by retraining
Published Oct 9
Evaluated models ResNet-32 Abstract Label 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.
Proposes the Cancerous backdoor, which progressively corrupts reasoning using invisible stage tokens.
Published Oct 9
Evaluated models Llama-3.1-8B-Instruct , Qwen2.5-7B-Instruct , Mistral-7B-Instruct-v0.3 Abstract Hidden 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.
提出 SLDR，用选择性层恢复与动态路由防御恶意微调
Published Oct 8
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。
提出热启动 Bandit 的部署前有界离线注入攻击
Published Oct 7
Evaluated models UCB , Thompson Sampling , ϵ-greedy 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 Oct 7
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。
提出 FAB 攻击，向声学基础模型植入可由环境声触发的后门
Published Oct 7
Evaluated models HuBERT-base , WavLM-base Abstract 论文提出了针对声学基模的 FAB 后门攻击，在弱威胁模型下实现了任务无关、物理可触发的后门植入。
提出 Safety-Aware Pruning，缓解视觉 token 剪枝放大的多模态越狱
Published Oct 7
Evaluated models LLaVA-1.5-7B Abstract 剪枝去掉背景 token 后注意力塌向恶意前景。这篇工作评估视觉 token 剪枝如何改变 VLM 在多模态越狱下的行为，并给出推理时防御 SAP。
提出持久视觉记忆注入，使对抗图像经缓存在触发时操控输出
Published Oct 7
Evaluated models Qwen3-8B , Qwen3-VL-8B-Instruct , Gemma-4-12B Abstract 论文提出 P-VMI，证实对抗输入被掩蔽后仍可经 KV 缓存持久控制模型，强调了保留状态的安全风险。
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
Published Oct 6
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。
测量带后门教师经 OPD 向干净学生迁移潜伏后门
Published Oct 6
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。 。
提出 TRACE，仅凭检查点更新审计有害合规微调目标
Published Oct 6
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。
提出 DP-MixMin，私有学习公开数据混合并做差分隐私微调
Published Oct 6
Evaluated models ResNet-18 , GPT-2 Medium Abstract DP-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。
构建多语言音视频 DeepFake 基准 BabelFake 并评测检测器
Published Oct 5
Evaluated models SpeechForensics , AVH-Align , (B)RAVEn Abstract BabelFake 为五语同意采集基准。BabelFake 是经同意与 IRB 批准采集的多语音视频 DeepFake 检测基准，用来把视觉操纵和语音合成拆开比较。
提出轨迹级可迁移对抗攻击，降低深度强化学习策略的期望回报
Published Oct 5
Evaluated models DQN , Double DQN (DDQN) Abstract 轨迹级后退视界攻击在 CartPole 上比逐步 FGSM 类和 Sign 噪声更压低回报。这项工作研究 DRL 上的迁移式黑盒观测攻击 ：受害参数不可访问时，在白盒代理上造扰动再迁移。
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 Oct 5
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。
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 THRIVE，为自回归 TTS 嵌入抗重建的多比特语音水印
Published Oct 3
Evaluated models Qwen3-TTS (0.6B) , Dots.tts (2B) 提出条件特征抑制防御，检出对抗补丁后抑制 VLA 内部特征
Published Oct 2
Evaluated models π0.5 , SmolVLA Abstract 条件抑制与 patch 相关的 SAE 特征，间歇攻击下可提高 SR，持续攻击下则不能。
提出 LiBRA，在潜空间双向优化以规避图像水印双侧检测
Published Oct 2
Evaluated models Stable Signature , YU1 , HiDDeN and others4 Abstract LiBRA 用双向潜空间优化规避双侧水印检测，并提高相对反转基线的保真。
揭示无触发虚假训练下事实回答与下游决策的审计差距
Published Oct 2
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。