Translation in progress Some entries still include text in the original language.

Oct 12 · Monday

All papers

Found 12 papers

Filter3

Attack type

Jailbreaks29 Prompt injection50 Poisoning and backdoors17 Model tampering12 Extraction and theft12 Detection evasion5

Defense type

Model hardening11 Inference-time intervention5 Permissions and approval1 Watermarking and provenance3

Impact

Data and privacy leaks11

System type

Models and APIs11

Attack surface / entry point

User input8 Training data3

Layer

Model layer3 Training and data layer8

System component

Multimodal3

Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 265papers have no tags yet and are not included in filtered results.

Attack type：Extraction and theft Defense type：Model hardening Defense type：Permissions and approval Clear all

Another 265papers do not have category tags yet and are not included in filtered results.

DefensesarXiv 2610.04426Oct 3
UnAct：无需梯度的定向激活干预实现类别遗忘
提出 UnAct，仅以前向激活干预从分类器中遗忘指定类别
arXiv
2610.04426
Published
Oct 3
Layer
Model layer
Setting
Models and APIs
Evaluated models
ResNet-18, ViT-B/16
- DefensesModel hardening
- AttacksExtraction and theft
AbstractUnAct 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。
Paper analysisRead analysis
DefensesarXiv 2609.40286Oct 1
研究者提出 COVER：在语言预算下选择源语言实现跨语言遗忘
提出 COVER，在语言预算下选择源语言实现跨语言知识遗忘
arXiv
2609.40286
Published
Oct 1
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Tiny Aya Global, Qwen3-4B-Instruct-2507, Llama-3.2-3B-Instruct
- DefensesModel hardening
- AttacksExtraction and theft
AbstractCOVER 在语言预算下选源语言，以降低未监督语言上相对 oracle 的残余访问。
语言预算下的多语言遗忘：遗忘监督只放在 K 种源语言上，目标是降低其余语言对同一事实的残余访问，而不是把每种评测语言都译成监督数据。
Paper analysisRead analysis
DefensesarXiv 2610.02418Oct 1
Whiteout：用混淆样本阻止 LLM 泄露个人敏感信息
提出 Whiteout，用伪装样本微调覆盖个人敏感信息关联
arXiv
2610.02418
Published
Oct 1
Layer
Training and data layer
Setting
Models and APIs
Attacker
Black-box, White-box
Evaluated models
Llama 3.2, Phi-3-mini, Gemma 3 and others7
- DefensesModel hardening
- AttacksExtraction and theft
Paper analysisRead analysis
DefensesarXiv 2609.37858Sep 29
存储不等于策略：面向 LLM 遗忘的状态条件支持控制
提出 Intervention Score 与 DIR-R，按干预效果选择并修订遗忘参数支持
arXiv
2609.37858
Published
Sep 29
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Llama 3 8B, Gemma 2 2B, Qwen2.5-7B and others4
- DefensesModel hardening
- AttacksExtraction and theft
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。
Paper analysisRead analysis
DefensesarXiv 2609.34591Sep 28
TULIP：按输入定位层级的定向大模型遗忘方法
提出 TULIP，按输入定位形成层并去除目标 token 对齐以实现定向遗忘
arXiv
2609.34591
Published
Sep 28
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Llama-3.1-8B, Llama-3.2-3B, Llama-3.2-1B and others7
- DefensesModel hardening
- AttacksExtraction and theft
AbstractTULIP 在形成结束的层去掉遗忘目标对齐，多数设置下 FQ 更高，4-bit 下 FQ 下降。
TULIP 是一种逐输入选择层的表示级遗忘方法，目标是去掉遗忘答案的形成、留下读出。输出级方法可能只压低特定回答的似然。RMU、LUNAR 等则对全部遗忘样本共用一层。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2609.34092Sep 28
机器遗忘在 ASR 中是否有效：arXiv 论文评估现有算法与评测工具
arXiv
2609.34092
Published
Sep 28
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
E-Branchformer
- DefensesModel hardening
- AttacksExtraction and theft
- ComponentAudio
Abstract梯度上升类方法更接近简单评测下的重训参照，复杂方法会被知情攻击认出，多说话人更差。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2609.33481Sep 27
IDUnlearn-Bench：面向视觉语言模型的个体级遗忘基准
提出 IDUnlearn-Bench，评测 VLM 个体级遗忘后的身份残留
arXiv
2609.33481
Published
Sep 27
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Qwen3-VL-2B, Qwen3-VL-4B, Qwen3-VL-8B and others6
- DefensesModel hardening
- AttacksExtraction and theft
- ComponentVision
AbstractIDUnlearn-Bench 用四族探针显示：压低直接属性访问后，身份仍常能被反向查出、绑定或重构。
IDUnlearn-Bench 评测 VLM 是否忘掉一个可跨模态识别的人，而不只是答不出某条监督事实。
Paper analysisRead analysis
DefensesarXiv 2609.19325Sep 17
AUDITPLAN：先提交安全计划再作答的可审计安全对齐方法
提出 AUDIT PLAN 先计划后作答，并用 FAITHGATE 绑定答案奖励与计划正确性
arXiv
2609.19325
Published
Sep 17
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Qwen2.5-1.5B-Instruct, Qwen2.5-3B-Instruct, Qwen-3-4B-Instruct and others4
- DefensesModel hardening
- AttacksJailbreaks
Abstract先承诺可机检安全计划，再让计划正确性决定能否拿到答案奖励。
AUDIT PLAN 把安全对齐从“只优化最终答案”改成“先提交隐藏的结构化安全计划，再按该计划作答”，并用 FAITHGATE 把高答案奖励限定在计划正确时。
Paper analysisRead analysis
DefensesarXiv 2609.15799Sep 15
RAPID：面向文生图服务的实时防未授权蒸馏防御
提出 RAPID，离线把抗蒸馏扰动写入文生图解码器
arXiv
2609.15799
Published
Sep 15
Layer
Model layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
PixArt, SSD, SDXL and others4
- DefensesModel hardening
- AttacksExtraction and theft
- ComponentImage generation
Paper analysisRead analysis
DefensesarXiv 2609.10608Sep 8
Adaptive Diffusion Freezing：抵御成员推理攻击的隐私保护扩散模型
提出 Adaptive Diffusion Freezing，降低扩散模型的成员推理风险
arXiv
2609.10608
Published
Sep 8
Layer
Training and data layer
Setting
Models and APIs
- DefensesModel hardening
- AttacksExtraction and theft
Abstract占位。
占位。
Paper analysisRead analysis
DefensesarXiv 2609.01677Sep 1
Skill-as-API：面向智能体软件工程的机密多智能体协作协议
提出 Skill-as-API，隔离多智能体协作中的 skill 正文与系统提示词
arXiv
2609.01677
Published
Sep 1
Layer
Application layer
Setting
Multi-agent systems, coding agent
- DefensesInstruction–data separation
- AttacksExtraction and theft
- ComponentMCP and skills
Abstract协议把 skill 正文留在所有者侧，四层限制发现内容、权限期限和注入拼接。
Skill-as-API 是面向智能体软件工程的协调协议，用来在跨组织调用时把 skill 的代码和系统提示词留在所有者进程。作者针对三条威胁：经通道或命名空间抽取专有方法，用注入让被调用模型输出系统提示词，以及协作结束后权限继续保留。
Paper analysisRead analysis
DefensesarXiv 2602.24210Feb 28
研究者提出 Staged Decoding 与 SFT 数据集，减少大推理模型推理轨迹中的隐私泄露
提出 Staged Decoding 与指令跟随训练，减少推理轨迹隐私泄露
arXiv
2602.24210
Published
Feb 28
Layer
Model layer
Setting
Models and APIs
Evaluated models
Qwen 3 1.7B, Qwen 3 4B, Qwen 3 8B and others6
- DefensesModel hardening
- AttacksExtraction and theft
- ComponentReasoning chains
Abstract用一般指令跟随训练加分段解码，可降低轨迹隐私泄露，但数学效用下降。
本文把大推理模型的隐私问题做成轨迹可控性：隐私指令是指令，轨迹不跟随就会把敏感上下文写进可被提示注入导出的文本。
Paper analysisRead analysis

You have reached the end

AI Safety Daily

Read the latest issueOne issue per day, updated as events develop