Translation in progress Some entries still include text in the original language.
Oct 12 · Monday
All papers Found 68 papers Another 232 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.
Propose KVeil, which adds perturbations to face videos to prevent unauthorized customization from reproducing identity
Published Oct 9
Evaluated models Wan2.2-TI2V-5B , HunyuanVideo-1.5 , Kling 3.0 and others4 Abstract KVeil adapts LETHE's reference-reading disruption to the challenge's two stages, ranking first on the leaderboard in both Final columns.
Proposes LETHE, which perturbs reference images to block unauthorized image-to-video
Published Oct 9
Evaluated models Wan2.1-I2V-14B , CogVideoX-5B-I2V , LightX2V and others6 Abstract LETHE suppresses the diversity of reference values and attracts attention, making protected images difficult to animate faithfully.LETHE is a protective perturbation added to personal images before publication, used to counter unauthorized image-to-video generation.
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 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。
固定有害指令,单轴归因图像到文本越狱的图像属性
Published Oct 4
Evaluated models GPT-4o , Claude Sonnet 4.5 , Qwen3-VL-8B-Instruct and others6 Abstract 固定文本的图像越狱拆解,只在 Qwen 上给出一条有界相关性归因。
遮挡审计 AdvWave-P 音频越狱的频带必要性与解码层关联
Published Oct 4
Evaluated models Qwen2-Audio-7B-Instruct , LLaMA-Omni Abstract AdvWave-P 的八频带排序与能量相关。这是对加性音频越狱 AdvWave-P 的频率与解码器深度审计,受害模型主要是 Qwen2-Audio-7B-Instruct,不提出新攻击。
提出 Bi-QSTO,优化能穿过质量筛选并在微调中削弱安全对齐的投毒样本
Published Oct 1
Evaluated models Llama2-7B-Chat , Llama3-8B-Instruct , Qwen2-7B-Instruct and others5 Abstract 质量筛选挡不住部分高分样本的安全削弱。Bi-QSTO 研究一个实际训练缺口:基于质量的数据选择会先于微调决定样本去留,而以往微调投毒多假设毒样本直接进入训练集。已对齐 LLM 的拒答会被少量有害或看似无害的微调样本削弱。作者用三个问题串起实验:选择能否排除毒样本。
提出只训练或篡改多智能体潜在通信链路来提高有害遵从
Published Sep 30
Evaluated models Llama-3.2-3B-Instruct , Qwen2.5-3B-Instruct , Gemma-3-1B-IT and others7 Abstract 潜在链路本身即可改变冻结智能体系统的安全,奖励既能攻击也能修复。学习得到的潜在通信链路会改变由冻结、对齐智能体组成的系统是否遵从有害请求。 。作者研究三种拓扑:双智能体、三级顺序协作,以及双专家加汇总。
提出 FDCU 双约束子空间投影遗忘,抵御再训练恢复有害知识
Published Sep 30
Evaluated models Qwen2.5-3B , Llama-3-8B-Instruct , Qwen-3-8B Abstract FDCU 用双掩码限制浅层抑制,在知识擦除与安全输出上提高再训练后的稳定性。FDCU 是一种约束参数更新的遗忘方法,用来降低再训练把已抑制恶意行为重新引出的可能。
提出 SceneJail,利用视频情境上下文越狱视频多模态模型
Published Sep 30
Evaluated models Qwen2.5-VL-7B-Instruct , Qwen2.5-VL-32B-Instruct , Qwen3-VL-8B-Instruct and others8 提出 VaccineBooster,在对齐阶段混合嵌入扰动与梯度衰减抵御有害微调
Published Sep 29
Evaluated models Llama-2-7B Abstract VaccineBooster 把两种对齐阶段扰动合在一步里,内容更安全但拒答保留更少。VaccineBooster 是一种对齐阶段防御:在提供方还看得到训练过程、还没接触用户微调数据时,把嵌入扰动和权重级梯度衰减合成同一次更新,用来应对之后不受信任的微调。有害微调可以只混入少量有害指令-回答对,就削弱拒答并提高有害输出的可能,任务效用却仍然好看。过滤会漏掉隐蔽样本,事后修复又往往不知道该修哪里。
检验层冻结与奇异方向截断能否挡住少样本有害微调
Published Sep 29
Evaluated models Llama-3.1-8B-Instruct , Llama-3.1-Tulu-3-8B-DPO , OLMo-2-1124-7B-Instruct and others10 Abstract 定位与恢复可以复现,但知情攻击能把损伤移出冻结带,并打败跨检查点的截断修复。这篇工作检验:把拒答定位到某些层或更新方向之后,能不能在攻击者知情时还守住拒答。 。
构建 TriggerBench 并训练 ThinkingGuard,检测图文隐式风险
Published Sep 29
Evaluated models ThinkingGuard(Qwen3-VL-8B-Instruct) Abstract TriggerBench 隔离触发条件,ThinkingGuard 用分步归因在隐式风险检测上取得作者报告的较高平均分。这篇工作研究多模态隐式风险检测 :图文各自良性,风险只在关键元素与触发元素交互时出现。
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 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 用后验重加权解释多模态上下文越狱,并提出风险门控的良性示例注入防御
Published Sep 8
Evaluated models Qwen3-VL-2B , Qwen3-VL-4B , Qwen3-VL-8B and others8 Abstract 后验重加权解释上下文越狱的缩放,并支持风险门控的良性示例防御。Zhang 等把多模态 ICL 越狱写成对齐模型在安全与有害隐模式之间的推理时后验漂移,并据此做风险门控防御。
提出 DIVA,将有害意图拆成视觉锚点与运动文本以越狱视频生成模型
Published Sep 7
Evaluated models Wan , CogVideoX , LTX-Video and others9 Abstract DIVA 用视觉锚点抑制语义漂移做视频越狱,并给出多条件安全基准 TI2VSafetyBench。DIVA 是针对带参考图的视频生成模型的免训练 black-box 越狱:用一致性机制本身堵住语义漂移这条安全逃逸路径。 。