Translation in progress Some entries still include text in the original language.
Oct 13 · Tuesday
All papers Found 18 papers Another 213 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.
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 Persist-3 序贯检验，从隐藏表示识别未见越狱源
Published Oct 7
Evaluated models Llama2-7B-chat , Qwen2.5-7B-Instruct , Gemma-2-9B-it and others4 Abstract Persist-3 用有限窗口检验攻击流是否超出已知越狱源，检出率低于全历史 SR-DKW，但决策只依赖最近观测。Persist-3 是一个有限记忆的序贯检测器，用来判断攻击表示流是否已经不能由任何一个已知越狱源解释。 。
评测文生图模型的有害内容生成能力与审核检测缺口
Published Oct 5
Evaluated models SDXL , Stable Diffusion 3.5 Large , FLUX.1-dev and others12 用分类后缀提升激活探针对分布外恶意输入的检测
Published Oct 2
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。
用能量势垒信号检测扩散语言模型的越狱
Published Sep 25
Evaluated models LLaDA-8B-Instruct , LLaDA-1.5 , Dream-v0-Instruct-7B and others4 Abstract 能量势垒把 dLLM 越狱分成掩盖初始状态或中途越垒，三个 logit 信号据此互补检测。Bach 等提出用去噪能量景观解释 masked diffusion 语言模型上的越狱，并由此得到三个免训练检测信号。 。
提出 AEGIS，用中层内部信号选择性激活后层适配器拦截音频越狱
Published Sep 24
Evaluated models Qwen2-Audio-7B , VITA-1.5 , Ultravox v0.5 and others6 提出 GUARD，在零样本 TTS 上用激活转向遗忘说话人身份以阻止再识别
Published Sep 23
Evaluated models CosyVoice2-0.5B , F5-TTS , FireRedTTS Abstract GUARD 用门控激活转向把退出说话人推向冒充相似度，以抑制再识别并保住音质。GUARD 用门控激活转向做说话人身份遗忘，使零样本 TTS 对退出说话人难以再识别。
提出 PALS，用潜空间平滑防御全双工语音对话的对抗扰动
Published Sep 23
Evaluated models Moshi-style agent 提出局部稀疏无监督安全检测，标记偏离安全分布的输入
Published Sep 17
Evaluated models qwen2-1.5b-instruct , ministral-8b-instruct , llama3-8b-instruct and others6 提出 SPARK，在预填充阶段修复多模态 KV 记忆以防御视觉语言模型越狱
Published Sep 13
Evaluated models LLaVA-OneVision-7B , Chameleon-7B , Qwen2-VL-7B and others4 提出 HERALD，用跨层有害投影轨迹检测越狱提示
Published Sep 12
Evaluated models Llama-3.1-8B-Instruct , Mistral-7B-Instruct , OLMo2-7B-Instruct and others4 Abstract HPD 用跨层投影形状区分有害与良性提示。
在写入点定位拒答方向，经刚性旋转跨架构迁移并门控放大
Published Sep 4
Evaluated models Falcon-Mamba-7B , Falcon3-Mamba-7B , Llama-3.1-8B and others8 Abstract 拒答方向可经刚性旋转跨架构迁移，但必须在各架构的写入点读取。作者研究拒答表示是否随架构改变，结论是表示共享、读取位置随架构变。
提出 EraseSAE，在文生视频模型中局部擦除指定概念
Published Sep 3
Evaluated models CogVideoX-5B , HunyuanVideo , Flux.1 [dev] Abstract EraseSAE 用单义特征局部擦除。EraseSAE 把 SAE 用于 DiT 文生视频模型的概念擦除。作者要解决的是：全局改权重或表面负提示对不齐细粒度、分布式的概念，而且会改到无关时空区域。
提出机器遗忘式生成图像检测，用遗忘前后特征相似度区分自然图与生成图
Published Aug 1
Evaluated models DINOv2 ViT-S/14 , DINOv2 ViT-B/14 , DINOv2 ViT-L/14 and others9 Abstract 作者用剪枝或 LoRA 遗忘拉开生成图与自然图的特征相似度，并在多个检测基准上报告结果。这篇工作把 AI 生成图像检测写成对预训练视觉模型的机器遗忘：原模型对两类图像损失都低，剪枝后生成图特征变得更不像原模型。
提出 D²-Monitor，按犹豫步数把难提示路由到更强探针
Published May 25
Evaluated models LLaDA-8B-Base , LLaDA-8B-Instruct , LLaDA-1.5-8B and others4 Proposes CASG to resolve direction conflicts in multi-category safety guidance for text-to-image generation.
Published Feb 24
Evaluated models Stable Diffusion v1.5 , SDv3 提出 LODO 评测恶意提示探针的跨数据集泛化
Published Feb 15
Evaluated models Llama-3.1-8B-Instruct , Gemma-3-27B , Qwen-3.5-4B and others5 Abstract 作者称 LODO 露出 CV 高估，数据集身份与安全方向部分重叠，标准域泛化干预关不上。LODO 用来检查恶意提示词分类器换成没见过的数据集后，还剩多少检测能力。 。
Proposes SafeThink, an inference-time defense that monitors the chain of thought and injects a redirect prefix when it falls below a threshold.
Published Feb 12
Evaluated models R1-Onevision-7B , OpenVLThinker-7B , VLAA-Thinker-7B and others6 You have reached the end