Translation in progress Some entries still include text in the original language.
Oct 13 · Tuesday
All papers Found 54 papers Another 203 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 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。
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。
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。
提出 SAGE，用少量已核验样本按相似度清洗投毒训练数据
Published Oct 1
Evaluated models ResNet-18 , VGG-16 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 ReGap，用自生成伪监督微调恢复模型中的隐私关联
Published Oct 1
Evaluated models GPT-2 Small , GPT-2 Medium , GPT-2 Large and others6 Abstract 提出无真实私有监督的恢复攻击 ReGap，实验显示微调可通过自生成数据唤醒模型潜在隐私关联，突显后适应隐私风险。
用零空间投影净化 LoRA 微调模型后门并保留基座与下游能力
Published Oct 1
Evaluated models LLaMA2-7B-Chat , LLaMA2-13B-Chat , Qwen2-7B-Instruct and others6 提出 GSU 在发布前封堵敏感门梯度，抵抗下游微调获取禁止能力
Published Sep 30
Evaluated models Llama-3.2-1B , Llama-3.2-3B , Llama-3.1-8B and others7 Abstract GSU 在发布前封堵 SiLU 门的获取梯度，并在 TOFU 与 WMDP 上降低下游微调得分。Gradient-Sealed Unlearning（GSU） 是一种发布前的 preemptive unlearning：不删除模型里已经有的能力，而是让指定禁止域能力更难被发布后的微调获取。
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 BadAction，通过动作触发器给交互式视频生成植入后门
Published Sep 30
Evaluated models Astra (Wan2.1-T2V-1.3B) 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 Sep 30
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。
提出 Intervention Score 与 DIR-R，按干预效果选择并修订遗忘参数支持
Published Sep 29
Evaluated models Llama 3 8B , Gemma 2 2B , Qwen2.5-7B and others4 Abstract 作者把局部化遗忘拆成初始干预选择和检查点修订，并在三个设置里分别给证据和边界。Storage Is Not Strategy 研究局部化 LLM 遗忘里两个常被混在一起的决定：一开始该更新哪些参数，以及优化改变模型之后这个选择要不要改。
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。