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

Oct 12 · Monday

All papers

Found 219 papers

Filter5

System component

MCP and skills82 Memory78 RAG50 Guardrails and judges86 Monitors81 Reasoning chains45 Permissions and sandboxing92 Fine-tuning and open weights14 Vision63 Audio21 Video5 Image generation22 Embeddings13 Multimodal103

Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 254papers have no tags yet and are not included in filtered results.

Layer：Model layer Layer：Application layer System component：Memory System component：Guardrails and judges System component：Vision Clear all

Another 254papers do not have category tags yet and are not included in filtered results.

Risky behaviorarXiv 2610.12822Oct 9
Reward hacking as a transmissible strategy in multi-agent coding systems
Show that reward hacking can propagate through code, memory, and plans in multi-agent coding systems
arXiv
2610.12822
Published
Oct 9
Layer
Application layer
Setting
Multi-agent systems, coding agent
Evaluated models
Qwen2.5-Coder-14B, Gemma-3-12B, Qwen2.5-Coder-7B and others16
- RiskReward hacking
- ComponentMemory
AbstractThe authors state that reward hacking can spread across models and affect routing and aggregation, and that review can reduce delivery.
Reward hacking induced by coding RL can leave the source agent and enter subsequent collaboration and system coordination.
Paper analysisRead analysis
AttacksarXiv 2610.12292Oct 9
The paper proposes an option channel attack: changing one option label can make an agent guardrail allow a prohibited operation
Proposing an option channel attack that renames option labels so that agent guardrails allow prohibited operations
arXiv
2610.12292
Published
Oct 9
Layer
Application layer
Setting
LLM applications
Attacker
Black-box
Evaluated models
LAYA-TD, LAYA-EN, LAYA-ML and others5
- AttacksDetection evasion
- ComponentGuardrails and judges
AbstractThe authors' evaluation indicates that typed models should not serve as final safety arbiters; they are susceptible to off-policy text and option name manipulation, and confidence filtering cannot resist attacks.
Paper analysisRead analysis
Analysis and theoryarXiv 2610.12083Oct 8
Paper audits LLM judge reliability: repeated agreement does not equal correctness
Auditing LLM judge reliability and defining trustworthy verdict rate and accuracy upper bound
arXiv
2610.12083
Published
Oct 8
Layer
Model layer
Setting
Models and APIs
Evaluated models
GPT-5.2, GPT-4o, Gemini-2.5-Pro and others6
- ComponentGuardrails and judges
AbstractThe authors claim that comparing judges should look at T, not just accuracy.
The authors audit whether LLM-as-a-Judge decisions can still be treated as actionable ground truth under repeated queries, swapped candidate order, and changed prompt formats.
Paper analysisRead analysis
InterpretabilityarXiv 2610.11986Oct 8
FearCaut-Qwen: Activation steering changes the risk judgment criteria of vision-language models
Using fear activation steering to test the hazard judgment criteria of vision-language models
arXiv
2610.11986
Published
Oct 8
Layer
Model layer
Setting
LLM applications
Evaluated models
Qwen/Qwen2.5-VL-7B-Instruct
- ComponentVision
Paper analysisRead analysis
DefensesarXiv 2610.11942Oct 8
LGWM: Verifying GUI agent screen transitions with action-conditioned world models
Proposes LGWM, using an action-conditioned world model to verify GUI agent screen transitions.
arXiv
2610.11942
Published
Oct 8
Layer
Application layer
Setting
Browser and computer use
Evaluated models
LGWM
- DefensesMonitoring and auditing
- AttacksPrompt injection
- ComponentMonitors
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.11773Oct 8
研究提出 Agent 安全新维度：识别未履行的安全义务
提出义务识别基准 ObligationBench，并训练护栏识别 Agent 未履行的安全义务
arXiv
2610.11773
Published
Oct 8
Layer
Application layer
Setting
coding agent
Evaluated models
ObligationGuard, Claude-Opus-4.8, GPT-5.6-Sol and others17
- DefensesDetection and filtering
- RiskDangerous agent actions
- ComponentGuardrails and judges
Abstract义务是护栏缺口。
作者认为编码智能体的安全不能只靠拦截禁止动作，还要找出终止时仍未做的安全关键动作，并将其称为义务。
Paper analysisRead analysis
DefensesarXiv 2610.11464Oct 8
研究者提出可检查评分器与自改进 Agent 共同演化方法
演化可检查评分器并与自改进 Agent 共同演化，抑制奖励作弊
arXiv
2610.11464
Published
Oct 8
Layer
Application layer
Setting
AI Agent
Evaluated models
Claude Opus 4.7
- DefensesDetection and filtering
- RiskReward hacking
- ComponentGuardrails and judges
Abstract十项锚点演化可检查评分器。
开放任务没有验证器时，被演化的是评分器，而不是再补一条手写 rubric。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.10947Oct 8
RH-Detect：统一的奖励作弊检测基准发布
构建统一奖励作弊检测基准 RH-Detect 并评测零样本检测器
arXiv
2610.10947
Published
Oct 8
Layer
Application layer
Setting
LLM applications
Evaluated models
Qwen 3.5 Flash, Gemini 3.1 Flash Lite, Grok 4.1 Fast and others7
- DefensesDetection and filtering
- RiskReward hacking
- ComponentGuardrails and judges
Paper analysisRead analysis
AttacksarXiv 2610.10742Oct 8
BRANCH 方法绕过 6 套多扫描器护栏系统
提出 BRANCH 方法，用分支搜索绕过多扫描器护栏
arXiv
2610.10742
Published
Oct 8
Layer
Application layer
Setting
LLM applications
Attacker
Black-box
Evaluated models
Meta Llama Guard 3 8B, Meta Llama Guard 7B, Meta Llama Guard 3 1B and others18
- AttacksDetection evasion
- TechniqueAutomated search
- ComponentGuardrails and judges
Paper analysisRead analysis
AttacksarXiv 2610.09920Oct 7
研究：多向量视觉文档索引可被反演还原页面内容
提出流匹配逆向框架，从多向量视觉文档索引重构页面
arXiv
2610.09920
Published
Oct 7
Layer
Application layer
Setting
LLM applications
Evaluated models
Tomoro-ColQwen3-8B, ColQwen3.5-4.5B, Tomoro-colqwen3-embed-4b and others13
- AttacksExtraction and theft
- ComponentEmbeddings
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
AttacksarXiv 2610.10612Oct 7
PyCache Trap 利用 Python 缓存替换绕过七款 Agent Skill 扫描器
提出 PyCache Trap，用字节码缓存替换绕过 Agent 技能扫描器
arXiv
2610.10612
Published
Oct 7
Layer
Application layer
Setting
coding agent
Evaluated models
CISCO, HSS, AUDITOR and others8
- DefensesDetection and filtering
- AttacksDetection evasion
- TechniqueDefense-targeted
- ComponentMCP and skills
Abstract论文剖析智能体技能审查与执行脱节盲区，提出 PyCache Trap 攻击并构建基于可信复现的 EAV 防御。
这篇论文研究了智能体技能（Agent Skills）安全扫描中的“审查–执行脱节”问题，并提出了对应的攻击验证与防御方案。
Paper analysisRead analysis
Analysis and theoryarXiv 2610.09517Oct 7
审计发现 CLIP 安全分数主要反映画面与提示词场景的相似度
开环审计冻结视觉语言安全分，发现其反映场景模板而非危险感知
arXiv
2610.09517
Published
Oct 7
Layer
Model layer
Setting
Embodied systems and robotics
Evaluated models
CLIP ViT-B/32, CLIP ViT-L/14, Qwen2-VL-7B
- ComponentVision
Abstract论文审计了冻结 VLM 安全评分机制，指出其反映场景模板而非危险感知，为安全强化学习评估提供了严谨基准。
本文是一项针对安全强化学习中冻结视觉语言模型（VLM）安全评分机制的实证审计研究。
Paper analysisRead analysis
DefensesarXiv 2610.09469Oct 7
Secure-CUA: Constraining the Impact of Untrusted Content in Computer-Use Agents
提出 Secure-CUA，用动作事务约束计算机使用智能体的不可信影响
arXiv
2610.09469
Published
Oct 7
Layer
Application layer
Setting
web agent
Evaluated models
Claude Opus 5, Gemini 3.8 Flash, GPT-5.6-Sol
- DefensesInstruction–data separation
- AttacksPrompt injection
- ComponentVision
AbstractSecure-CUA 用每步事务约束不可信影响，良性 TSR 接近 Vanilla-CUA。
Secure-CUA 要限制图形界面里不可信内容对 CUA 决策和 GUI 执行的影响，同时保留合法任务所需的不可信信息。
Paper analysisRead analysis
AttacksarXiv 2610.09240Oct 7
WebMirage hijacks visual web agents' browser operations with adversarial images
提出 WebMirage，用局部对抗图像劫持视觉网页 Agent 的浏览器操作
arXiv
2610.09240
Published
Oct 7
Layer
Application layer
Setting
web agent
Evaluated models
LLaVA-v1.5-13B, LLaVA-v1.6-34B, MiniCPM-o-8B and others6
- AttacksPrompt injection
- TechniqueGradient and representation optimization
- ComponentVision
Abstract论文提出了针对网络智能体的端到端视觉红队框架 WebMirage，揭示了从视觉定位到浏览器执行的安全脆弱性。
WebMirage 是一个面向视觉定位网络智能体的流水线感知端到端红队攻击框架。
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
Evaluation and benchmarksarXiv 2610.08902Oct 7
论文提出 agent plasticity 指标衡量智能体通过经验自我改进的效率
提出 agent plasticity 指标衡量智能体通过经验自我改进的效率
arXiv
2610.08902
Published
Oct 7
Layer
Application layer
Setting
coding agent
Evaluated models
Claude Fable 5, Claude Opus 5, GPT-5.6 Sol and others10
- ComponentMemory
Abstract作者用固定权重协议比较前沿模型把经验写成制品后的留出增益与习得效率。
作者用 agent plasticity 衡量一件事：权重冻结时，智能体把经验写成可继承制品，再换成留出能力增益，这个转换有多省。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.08571Oct 6
RAG-PIBench：可信 RAG 系统中提示注入检测的泄漏感知基准
提出 RAG-PIBench，评测 RAG 上下文中提示注入检测器
arXiv
2610.08571
Published
Oct 6
Layer
Application layer
Setting
LLM applications
Evaluated models
DistilBERT-base-uncased, DeBERTa-v3-base, MiniLM
- DefensesDetection and filtering
- AttacksPrompt injection
- ComponentRAG
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.08215Oct 6
Learn2Play Bench 发布：用规则陌生的文本游戏评测 LLM Agent 从经验中学习的能力
用 Learn2Play Bench 评测 Agent 从经验中学习陌生规则
arXiv
2610.08215
Published
Oct 6
Layer
Application layer
Setting
AI Agent
Evaluated models
Claude Opus 5.5, GPT-6 Astra, Claude Opus 5 and others11
- ComponentMemory
Abstract用隐藏规则游戏把新学习与预训练知识分开，并比较模型、记忆方式和人类。
Learn2Play Bench 用新设计的隐藏规则文本游戏，评测 LLM 智能体在不更新权重时从交互中学习的能力。
Paper analysisRead analysis
OtherarXiv 2610.08048Oct 6
DAEDALUS：用自生成任务自举 Agent 记忆
提出 DAEDALUS，用自生成任务自举 Agent 记忆
arXiv
2610.08048
Published
Oct 6
Layer
Application layer
Setting
AI Agent
Evaluated models
GPT-5.4-mini, GPT-5.6 Luna, Qwen3.6-35B-A3B and others11
- ComponentMemory
Paper analysisRead analysis

AI Safety Daily

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

All papers

Attack type

Attack technique

Model risks

Defense type

Impact

System type

Attacker access

Attack surface / entry point

Layer

System component

Reward hacking as a transmissible strategy in multi-agent coding systems

The paper proposes an option channel attack: changing one option label can make an agent guardrail allow a prohibited operation

Paper audits LLM judge reliability: repeated agreement does not equal correctness

FearCaut-Qwen: Activation steering changes the risk judgment criteria of vision-language models

LGWM: Verifying GUI agent screen transitions with action-conditioned world models

研究提出 Agent 安全新维度：识别未履行的安全义务

研究者提出可检查评分器与自改进 Agent 共同演化方法

RH-Detect：统一的奖励作弊检测基准发布

BRANCH 方法绕过 6 套多扫描器护栏系统

研究：多向量视觉文档索引可被反演还原页面内容

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

PyCache Trap 利用 Python 缓存替换绕过七款 Agent Skill 扫描器

审计发现 CLIP 安全分数主要反映画面与提示词场景的相似度

Secure-CUA: Constraining the Impact of Untrusted Content in Computer-Use Agents

WebMirage hijacks visual web agents' browser operations with adversarial images

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

论文提出 agent plasticity 指标衡量智能体通过经验自我改进的效率

RAG-PIBench：可信 RAG 系统中提示注入检测的泄漏感知基准

Learn2Play Bench 发布：用规则陌生的文本游戏评测 LLM Agent 从经验中学习的能力

DAEDALUS：用自生成任务自举 Agent 记忆