Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 215papers have no tags yet and are not included in filtered results.

Attack type：Jailbreaks Attack type：Extraction and theft Attack surface / entry point：Documents and messages Attack surface / entry point：Tool and skill descriptions Attack surface / entry point：Training data Clear all

Another 215papers do not have category tags yet and are not included in filtered results.

DefensesarXiv 2610.06636Oct 6
差分隐私混合公共数据集提升隐私学习效果
提出 DP-MixMin，私有学习公开数据混合并做差分隐私微调
arXiv
2610.06636
Published
Oct 6
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
ResNet-18, GPT-2 Medium
- AttacksExtraction and theft
AbstractDP-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。
Paper analysisRead analysis
AttacksarXiv 2610.05943Oct 5
CRIME 框架利用良性 Agent 技能组合诱发恶意行为
提出 CRIME 框架，组合良性 Agent 技能诱发恶意行为
arXiv
2610.05943
Published
Oct 5
Layer
Application layer
Setting
AI Agent
Evaluated models
DeepSeek-V4-Flash, GPT-4o-mini, GPT-4o and others6
- AttacksJailbreaks
- TechniqueAutomated search
- ComponentMCP and skills
Abstract良性技能经 DCA 或 ACA 组合后可产生恶意环境后果，单独审查对不上。
CRIME 把公共技能库中各自通过审查的技能配成两两组合，看联合执行能否实现指定恶意任务。。
Paper analysisRead analysis
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
AttacksarXiv 2610.01365Oct 1
ReGap：无需真实隐私数据即可通过微调重新提取模型隐私关联
提出 ReGap，用自生成伪监督微调恢复模型中的隐私关联
arXiv
2610.01365
Published
Oct 1
Layer
Training and data layer
Setting
Models and APIs
Attacker
White-box, Gray-box
Evaluated models
GPT-2 Small, GPT-2 Medium, GPT-2 Large and others6
- AttacksExtraction and theft
- ComponentFine-tuning and open weights
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。
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
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
AttacksarXiv 2609.33985Sep 28
研究：众包微调数据投毒可放大专有指令抽取
提出主题锚点投毒，放大众包微调后的专有指令提取
arXiv
2609.33985
Published
Sep 28
Layer
Training and data layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
Qwen2.5-7B-Instruct, Qwen2.5-14B-Instruct, Llama-3.1-8B-Instruct and others4
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentFine-tuning and open weights
Abstract论文提出基于主题锚点的数据中毒攻击，证明低比例的中毒样本可在黑盒条件下隐蔽放大微调指令泄漏。
Paper analysisRead analysis
DefensesarXiv 2610.00209Sep 21
基于差分隐私扩散模型的能源时间序列插补：裁剪感知的目标条件化方法
提出时间步感知目标条件化，改善差分隐私能源时序插补效用
arXiv
2610.00209
Published
Sep 21
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
DiT-style Transformer
- AttacksExtraction and theft
Abstract固定阈值 DP-SGD 下，v-prediction 加 ᾱt 加权降低能源时序插补误差，并降低后期低 SNR 的裁剪前梯度。
作者研究固定阈值 DP-SGD 训练的扩散模型，如何在余弦调度的后期低 SNR 时间步避免 ε-prediction 产生易被裁剪的大梯度，从而做能源时序缺失值恢复。
Paper analysisRead analysis
DefensesarXiv 2609.05903Sep 5
EvoSafeHarness：为 Agent 自动演化模型与领域专属安全护栏
为冻结模型与目标域自动搜索可部署的 Agent 安全 harness
arXiv
2609.05903
Published
Sep 5
Layer
Application layer
Setting
AI Agent
Evaluated models
Sonnet 4.6, GLM-5, Kimi-K2.5 and others11
- DefensesInstruction–data separation
- AttacksPrompt injection
- ComponentPermissions and sandboxing
Paper analysisRead analysis
DefensesarXiv 2609.02293Sep 2
SEAL：通过共享专家对齐强化 MoE 全局安全
提出 SEAL，在 MoE 共享专家上做 DPO-LoRA 对齐以抵御越狱与权重篡改
arXiv
2609.02293
Published
Sep 2
Layer
Training and data layer
Setting
Models and APIs
Attacker
White-box, Black-box
Evaluated models
Qwen1.5-MoE-A2.7B, DeepSeekMoE-16B, GLM-4.7-Flash and others4
- DefensesModel hardening
- AttacksModel tampering
AbstractSEAL 把 DPO-LoRA 限制在混合 MoE 共享专家上，SEAL++再保护已有安全神经元方向。
SEAL是面向混合 MoE 的训练时防御：只对齐始终执行的共享专家，使安全不依赖路由器是否把 token 送到安全专家。
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
AttacksarXiv 2607.26115Jul 29
OpenAI releases GPT-Red: training automated red-teaming agents through self-play
训练自博弈红队智能体 GPT-Red，规模化生成提示注入与越狱攻击
arXiv
2607.26115
Published
Jul 29
Layer
Application layer
Setting
AI Agent
Attacker
Black-box
Evaluated models
GPT-5.6, GPT-5.5, GPT-5.4 and others8
- AttacksPrompt injection
- TechniqueAutomated search
Paper analysisRead analysis
AttacksarXiv 2607.00481Jul 1
研究者提出 SMT 框架，通过模拟审核轨迹越狱函数调用 LLM
提出 SMT 框架，用模拟审核轨迹越狱函数调用 LLM
arXiv
2607.00481
Published
Jul 1
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box, White-box
Evaluated models
GPT-4o, GPT-5.4, Qwen3-Max and others7
- AttacksJailbreaks
- TechniqueMulti-turn escalation
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2602.14161Feb 15
When benchmarks lie: evaluating malicious prompt classifiers under real-world distribution shift
提出 LODO 评测恶意提示探针的跨数据集泛化
arXiv
2602.14161
Published
Feb 15
Layer
Model layer
Setting
Models and APIs
Evaluated models
Llama-3.1-8B-Instruct, Gemma-3-27B, Qwen-3.5-4B and others5
- DefensesMonitoring and auditing
- AttacksPrompt injection
- ComponentMonitors
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。。
Paper analysisRead analysis
AttacksarXiv 2601.22169Jan 19
In Vino Veritas and Vulnerabilities：用醉酒语言诱导检验 LLM 安全
用醉酒语言诱导改编对话模型并评测越狱与隐私泄露
arXiv
2601.22169
Published
Jan 19
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
LLaMA2-7B, LLaMA3-8B, Mistral-7B and others5
- AttacksJailbreaks
- TechniqueRole-play and persuasion
- ComponentFine-tuning and open weights
Abstract醉酒语言诱导在五个 LLM 上抬高越狱与部分隐私错误，作者将其视为攻击向量。
作者把醉酒语言当作攻击向量，检验把 LLM 适配成醉酒发短信的模式后，越狱和情境隐私是否变差。动机是人类醉酒时的失范与泄密，以及 LLM 模拟人设的能力。
Paper analysisRead analysis
DefensesarXiv 2601.09292Jan 14
Blue Team Testing Function-Calling Agents
提出 Tool Obfuscation 并评测函数调用 Agent 的攻防
arXiv
2601.09292
Published
Jan 14
Layer
Application layer
Setting
AI Agent
Evaluated models
Qwen3:8B, Llama-3.2:3B, Granite3.2:8B and others5
- DefensesDetection and filtering
- AttacksPrompt injection
- ComponentGuardrails and judges
Paper analysisRead analysis
AttacksarXiv 2601.05466Jan 9
Jailbreaking large language models through reinforcement learning with iterative tool disguise attacks
提出 iMIST，用工具伪装和 PPO 迭代越狱大语言模型
arXiv
2601.05466
Published
Jan 9
Layer
Application layer
Setting
AI Agent
Attacker
Black-box
Evaluated models
DeepSeek-V3, Qwen3-32B, GPT-OSS-120B
- AttacksJailbreaks
- TechniqueAutomated search
AbstractiMIST 用工具伪装和在线 PPO 做黑盒越狱，Table I 中 JADES 高于同格基线。
iMIST是黑盒越狱方法：把恶意查询伪装成合法工具调用，再用在线强化学习把响应的有害性分数往上推。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2510.14207Oct 16, 2025
论文提出多轮在线骚扰 Agent 基准
提出多轮在线骚扰 Agent 基准，评测记忆、规划与微调越狱
arXiv
2510.14207
Published
Oct 16, 2025
Layer
Application layer
Setting
Multi-agent systems
Evaluated models
LLaMA-3.1-8B-Instruct, Qwen3-30B-A3B-Instruct, DeepSeekMoE-16B and others4
- AttacksJailbreaks
- ComponentFine-tuning and open weights
Paper analysisRead analysis

You have reached the end

AI Safety Daily

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