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

Oct 13 · Tuesday

All papers

Found 87 papers

Filter6

System component

MCP and skills4 Guardrails and judges5 Monitors32 Reasoning chains6 Permissions and sandboxing12 Fine-tuning and open weights4 Vision9 Audio4 Image generation4 Multimodal15

Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 201papers have no tags yet and are not included in filtered results.

Defense type：Inference-time intervention Defense type：Permissions and approval Defense type：Monitoring and auditing System type：Models and APIs System type：coding agent System type：web agent Clear all

Another 201papers do not have category tags yet and are not included in filtered results.

DefensesarXiv 2610.11012Oct 8
Redwood Research proposes the distillation double bind: using distillation to make models confess hidden behavior or inherit only capabilities
提出 DFI 与 DFC 蒸馏，定罪隐瞒失配的教师或安全提取能力
arXiv
2610.11012
Published
Oct 8
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Llama-3.3-70B-Instruct (AuditBench SFT/KTO organisms), Llama-3.3-70B-Instruct (DFI distilled student), Llama-3.1-8B-Instruct (DFI distilled student) and others4
- DefensesMonitoring and auditing
- RiskDeception and scheming
Abstract概括了论文针对隐瞒失配模型提出的 DFI 与 DFC 机制、主实验发现及适用边界。
Paper analysisRead analysis
DefensesarXiv 2610.10655Oct 8
Nullify：面向 LLM 遗忘的零空间激活引导免训练方法
提出零空间激活引导 Nullify，免训练擦除指定记忆并保持效用
arXiv
2610.10655
Published
Oct 8
Layer
Model layer
Setting
Models and APIs
Evaluated models
Llama-2-7B-chat-hf, Qwen3-8B
- DefensesInference-time intervention
- AttacksExtraction and theft
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.10276Oct 7
PatchBench 提出激活补丁局部评测协议，揭示越狱修复的附带损伤
提出 PatchBench-Local 协议，评测激活补丁修复越狱的局部附带损伤
arXiv
2610.10276
Published
Oct 7
Layer
Model layer
Setting
Models and APIs
Evaluated models
Qwen2.5-3B-Instruct, Qwen2.5-14B-Instruct, gemma-3-4b-it and others8
- DefensesInference-time intervention
- AttacksJailbreaks
Abstract论文提出 PatchBench 与局部评估协议，揭示了激活补丁在全局能力无损下仍存局部良性退化。
Paper analysisRead analysis
DefensesarXiv 2610.09703Oct 7
研究揭示 Token-Pruning 对 VLM 安全性的影响并提出 SAP 缓解机制
提出 Safety-Aware Pruning，缓解视觉 token 剪枝放大的多模态越狱
arXiv
2610.09703
Published
Oct 7
Layer
Model layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
LLaVA-1.5-7B
- DefensesInference-time intervention
- AttacksJailbreaks
- ComponentVision
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。
Paper analysisRead analysis
DefensesarXiv 2610.09167Oct 7
Persist-3 用顺序检验识别未见越狱来源，提升 MTK 与 Prompt Guard 2 防御效果
提出 Persist-3 序贯检验，从隐藏表示识别未见越狱源
arXiv
2610.09167
Published
Oct 7
Layer
Model layer
Setting
Models and APIs
Evaluated models
Llama2-7B-chat, Qwen2.5-7B-Instruct, Gemma-2-9B-it and others4
- DefensesMonitoring and auditing
- AttacksJailbreaks
- ComponentMonitors
AbstractPersist-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。。
Paper analysisRead analysis
InterpretabilityarXiv 2610.08559Oct 6
Study: Debiasing directions in LLM latent space primarily encode confidence rather than fairness
分析 LLM 去偏激活方向，发现其编码置信度而非公平性
arXiv
2610.08559
Published
Oct 6
Layer
Model layer
Setting
Models and APIs
Evaluated models
Llama-3.1-8B, Llama-3.1-8B-Instruct, Falcon3-7B-Base and others8
- DefensesInference-time intervention
- RiskBias and discrimination
Abstract论文揭示激活引导去偏见方向主要编码模型置信度而非公平性，偏见评分下降实为诱发弃答或使概率均分的假象。
Paper analysisRead analysis
DefensesarXiv 2610.07774Oct 6
研究者提出读取路由 logits 的 MoE 视觉语言模型多模态安全检测器
AI summary研究者提出一种轻量级路由 logits 安全检测器，在提示词预填充阶段读取 MoE 视觉语言模型的专家路由信号，在生成前判断多模态请求是否安全，且不修改模型参数或专家路由。
arXiv
2610.07774
Published
Oct 6
Layer
Model layer
Setting
Models and APIs
Evaluated models
Qwen3-VL-30B-A3B-Instruct, Kimi-VL-A3B-Instruct
- DefensesMonitoring and auditing
- RiskInappropriate refusal
- ComponentMonitors
Abstract读出而非操纵 router logits，可在不改模型的情况下降低两模型的组合式安全错误。
作者在两个指令微调 MoE VLM 上比较改模型与读路由。组合式风险按 HoliSafe 分成五类：图文各自可安全，合在一起才不安全，或不安全内容只在图像里。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.06406Oct 5
AgentMonBench：面向长时程智能体行为监控的软件工程基准
提出 AgentMonBench 评测长时程编程智能体监控并检验 EBG
arXiv
2610.06406
Published
Oct 5
Layer
Application layer
Setting
coding agent
Evaluated models
GPT-5.6 Luna, GPT-5.6 Terra, GPT-5.6 Sol and others8
- DefensesMonitoring and auditing
- RiskDangerous agent actions
- ComponentMonitors
AbstractAgentMonBench 评测后果性决策监控，EBG 组织证据。
长时程软件工程任务里，用户要判断智能体哪些选择值得核验。作者用 AgentMonBench 评测监控器，并用免训练的 EBG 组织带源证据。
Paper analysisRead analysis
DefensesarXiv 2610.06383Oct 5
基于影响力的曲率感知数据加权激活引导方法
提出曲率感知数据加权的激活转向方法，用于毒性抑制与概念控制
arXiv
2610.06383
Published
Oct 5
Layer
Model layer
Setting
Models and APIs
Evaluated models
Gemma-2-2B, Llama-3.2-3B
- DefensesInference-time intervention
Abstract影响加权传输压低毒性率，闭式 IF-Act 常换来更高困惑度与更低 MMLU。
推理时的影响加权激活传输，用曲率修正的样本权重替换传输转向里的均匀类均值。要做的是概念控制，而不是再训练。作者认为类均值被无关概念、噪声和少数 token 主导，编码的是词元级而非主题。
Paper analysisRead analysis
DefensesarXiv 2610.05246Oct 4
研究测量策略网关跨平台命令裁决的能力边界
提出策略门裁决已实现动作，并测量跨平台命令拦截边界
arXiv
2610.05246
Published
Oct 4
Layer
Application layer
Setting
coding agent
Evaluated models
claude-opus-5, deepseek-v4.1-flash, gpt-6-astra
- DefensesPermissions and approval
- RiskDangerous agent actions
- ComponentPermissions and sandboxing
Paper analysisRead analysis
DefensesarXiv 2610.05241Oct 4
StateWise: Diagnosing and repairing persistent operation state before agent actions
Propose StateWise, which diagnoses and repairs invalid persistent state before coding agent actions
arXiv
2610.05241
Published
Oct 4
Layer
Application layer
Setting
coding agent
Evaluated models
GPT-5.6-sol, DeepSeek v4 Pro, Qwen 3.7 Max and others4
- DefensesPermissions and approval
- RiskDangerous agent actions
- ComponentMemory
AbstractStateWise repairs stale persistent premises before action.
StateWise is a framework that diagnoses and repairs the persistent operational state of coding agents before executing actions with side effects.
Paper analysisRead analysis
DefensesarXiv 2610.05163Oct 4
Researchers construct segmented prompt injection on Claude Code and Codex and release a boundary action audit benchmark
提出 PAA 方法，在边界动作前审计长流程智能体的分阶段提示注入
arXiv
2610.05163
Published
Oct 4
Layer
Application layer
Setting
coding agent
Evaluated models
Claude Code (v2.1.172, Claude Sonnet 5), Codex CLI (v0.144.0, GPT-5.5)
- DefensesMonitoring and auditing
- AttacksPrompt injection
- ComponentMonitors
Abstract论文针对长流程注入提出边界动作审计与 PAA，在保持低误阻的同时实现高拦截率。
本研究针对长流程智能体在执行任务时面临的分阶段间接提示注入威胁，提出了在动作生效前进行拦截的边界动作审计框架与路径对齐归因方法。
Paper analysisRead analysis
DefensesarXiv 2610.04998Oct 4
EmoRSS：缓解大语言模型由情绪引发的过度拒答
提出 EmoRSS，推理时在拒绝子空间反向抵消情绪激活偏移以缓解过度拒答
arXiv
2610.04998
Published
Oct 4
Layer
Model layer
Setting
Models and APIs
Evaluated models
Llama-3.1-8B-it, Qwen3-8B
- DefensesInference-time intervention
- RiskInappropriate refusal
Paper analysisRead analysis
DefensesarXiv 2610.04967Oct 4
One Token Can Be Enough：用 Prefix Steering 连接提示与激活引导
提出 Prefix Steering，以短前缀激活引导控制行为并保住能力
arXiv
2610.04967
Published
Oct 4
Layer
Model layer
Setting
Models and APIs
Evaluated models
OLMo 3 7B, OLMo 3 32B, Qwen3 1.7B and others4
- DefensesInference-time intervention
- RiskInappropriate refusal
Abstract短前缀 steering 用已有方向做短暂干预。
本文用注意力层匹配连接提示词与激活干预，并把干预改成从末提示词 token 开始的短前缀。
Paper analysisRead analysis
DefensesarXiv 2610.04470Oct 3
SENTINEL：通过意图感知的输入-输出匹配防御 LLM 越狱
提出生成时防御 SENTINEL，匹配输入输出以抽出越狱意图并停写
arXiv
2610.04470
Published
Oct 3
Layer
Model layer
Setting
Models and APIs
Attacker
White-box
Evaluated models
Llama2-7B, Llama3-8B, Mistral-7B-v0.2 and others6
- DefensesInference-time intervention
- AttacksJailbreaks
Paper analysisRead analysis
DefensesarXiv 2610.02999Oct 2
OmniConfess：用 Token 级「忏悔」缓解全模态幻觉
提出 OmniConfess，冻结候选并按通道重打分以缓解全模态幻觉
arXiv
2610.02999
Published
Oct 2
Layer
Model layer
Setting
Models and APIs
Evaluated models
Qwen2.5-Omni-7B, Qwen3-Omni-30B-A3B, Nemotron-3-Nano-Omni-30B
- DefensesInference-time intervention
- RiskHallucination and fabrication
- ComponentVision
Paper analysisRead analysis
DefensesarXiv 2610.02844Oct 2
DNAlign：面向 LLM 的动态零空间安全对齐框架
提出 DNAlign，在不改权重的推理期用零空间约束控制信号做安全对齐
arXiv
2610.02844
Published
Oct 2
Layer
Model layer
Setting
Models and APIs
Evaluated models
Vicuna-7B, Qwen-7B
- DefensesInference-time intervention
- RiskInappropriate refusal
AbstractDNAlign 用零空间约束的推理期控制做安全对齐，Table I 平均奖励高于所列基线。
DNAlign 是一个不改 LLM 参数的安全对齐框架，用来在推理时压低有害输出，并减少对中性知识所支持的流畅、事实回答的改动。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.02741Oct 2
研究系统评估循环语言模型的思维链可监控性
系统评估循环语言模型加深 loop 后的 CoT 可监控性
arXiv
2610.02741
Published
Oct 2
Layer
Model layer
Setting
Models and APIs
Evaluated models
Ouro-1.4B-Thinking, Ouro-2.6B-Thinking, Nanbeige4.2-3B and others4
- DefensesMonitoring and auditing
- RiskDeception and scheming
- ComponentReasoning chains
Abstract加深 loop 可在部分压力测试任务降低可监测性，循环架构本身并不必然更难监测。
这项工作评估 LoopLM 的 CoT 可监测性：加深共享层的重复次数，以及“有循环结构”本身，会不会让决策关键因素更少出现在可监测文本里。
Paper analysisRead analysis
DefensesarXiv 2610.02557Oct 2
新 AI 辩论协议实现实例级最优：最坏情况正确且诚实为占优策略
提出 FBS Debate 协议，使诚实为占优策略且最坏情况正确
arXiv
2610.02557
Published
Oct 2
Layer
Training and data layer
Setting
Models and APIs
- DefensesMonitoring and auditing
- RiskDeception and scheming
AbstractFBS 辩论用 fbs 约束敏感分解，给出最坏情况占优策略，并匹配黑盒查询下界。
FBS Debate 是一套可扩展监督的辩论规则：两名计算方争论一个可递归分解的是非问题，人类只在最后选出的子问题上做黑盒判断。要处理的缺口是，prover-estimator debate 虽能覆盖稳定分解，但只在输入分布的平均意义下成立，而且 Bob 的获胜策略一般要跟着 Alice 的作弊策略变。作者要在同一问题类上，让每个错误实例都能被抓住，并让诚实且正确成为双方的占优策略。
Paper analysisRead analysis
DefensesarXiv 2610.02456Oct 2
SideKernel：面向 macOS 上 AI 编码 Agent 的 microVM 沙箱
为 macOS 编程 Agent 构建可用的 microVM 沙箱 SideKernel
arXiv
2610.02456
Published
Oct 2
Layer
Application layer
Setting
coding agent
- DefensesPermissions and approval
- RiskDangerous agent actions
- ComponentPermissions and sandboxing
Paper analysisRead analysis

AI Safety Daily

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

All papers

Attack type

Model risks

Defense type

Impact

System type

Attacker access

Attack surface / entry point

Layer

System component

Redwood Research proposes the distillation double bind: using distillation to make models confess hidden behavior or inherit only capabilities

Nullify：面向 LLM 遗忘的零空间激活引导免训练方法

PatchBench 提出激活补丁局部评测协议，揭示越狱修复的附带损伤

研究揭示 Token-Pruning 对 VLM 安全性的影响并提出 SAP 缓解机制

Persist-3 用顺序检验识别未见越狱来源，提升 MTK 与 Prompt Guard 2 防御效果

Study: Debiasing directions in LLM latent space primarily encode confidence rather than fairness

研究者提出读取路由 logits 的 MoE 视觉语言模型多模态安全检测器

AgentMonBench：面向长时程智能体行为监控的软件工程基准

基于影响力的曲率感知数据加权激活引导方法

研究测量策略网关跨平台命令裁决的能力边界

StateWise: Diagnosing and repairing persistent operation state before agent actions

Researchers construct segmented prompt injection on Claude Code and Codex and release a boundary action audit benchmark

EmoRSS：缓解大语言模型由情绪引发的过度拒答

One Token Can Be Enough：用 Prefix Steering 连接提示与激活引导

SENTINEL：通过意图感知的输入-输出匹配防御 LLM 越狱

OmniConfess：用 Token 级「忏悔」缓解全模态幻觉

DNAlign：面向 LLM 的动态零空间安全对齐框架

研究系统评估循环语言模型的思维链可监控性

新 AI 辩论协议实现实例级最优：最坏情况正确且诚实为占优策略

SideKernel：面向 macOS 上 AI 编码 Agent 的 microVM 沙箱