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.

Impact：Data and privacy leaks System type：Models and APIs System type：coding agent System type：web agent Layer：Application layer Layer：Infrastructure layer Clear all

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

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
Evaluation and benchmarksarXiv 2610.08871Oct 6
CredLeak-Bench
提出 CredLeak-Bench，评测 Agent 面对钓鱼的凭据泄露与恢复
arXiv
2610.08871
Published
Oct 6
Layer
Application layer
Setting
web agent
Evaluated models
Claude Sonnet 5, Claude Opus 4.8, Gemini 3.6 Flash and others7
- AttacksPrompt injection
Paper analysisRead analysis
AttacksarXiv 2610.07510Oct 6
PersistBD: attackers can strengthen backdoors before release so that they maintain high attack success rates even after developer SFT and RL.
提出 PERSISTBD，发布前强化供应链后门使其经良性 SFT 与 RL 后仍外发凭据
arXiv
2610.07510
Published
Oct 6
Layer
Application layer
Setting
coding agent
Evaluated models
Qwen2.5-Coder-3B-Instruct, Qwen2.5-Coder-7B-Instruct, Qwen3-Coder-30B-A3B-Instruct
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentFine-tuning and open weights
Abstract总结了论文关于后门在 SFT/RL 中存活的发现、两因子机制、PERSISTBD 方法与核心数字。
本文研究第三方 LLM 智能体在开发者进行良性监督微调（SFT）和强化学习（RL）后训练时的后门持久性风险。
Paper analysisRead analysis
AttacksarXiv 2610.06848Oct 6
TranScope：利用 CPU 性能计数器对 Transformer 做成员推断攻击
提出 TRANSCOPE，用 CPU 硬件计数器推断训练集成员
arXiv
2610.06848
Published
Oct 6
Layer
Infrastructure layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
BERT (4.4M-340M), ViT (22M-307M), Pythia (70M-2.8B) and others5
- AttacksExtraction and theft
- TechniqueSide channels
AbstractTRANSCOPE 揭示了分词器在嵌入表引发的硬件访问局部性差异，证实微架构计数器可突破恒定时间与置信度遮蔽防御。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.05586Oct 5
AgentDoxx：联网搜索 Agent 对匿名访谈文本的再识别评测
构建 AGENTDOXX 基准，评测联网搜索 Agent 对匿名访谈的再识别
arXiv
2610.05586
Published
Oct 5
Layer
Application layer
Setting
web agent
Evaluated models
GPT-5.6 Terra, GPT-5 Mini, GPT-4.1 Mini and others10
- AttacksMalicious use
- ComponentRAG
Abstract论文提出了已知身份的重识别基准 AGENTDOXX，揭示搜索与记忆双重风险及现有防御在检索轨迹中的泄露漏洞。
Paper analysisRead analysis
AttacksarXiv 2610.01768Oct 1
LLMLeak：借 LLM 合法网页抓取实施隐蔽数据外泄
提出 LLMLeak，借合法网页抓取把 URL 中的秘密隐蔽外传
arXiv
2610.01768
Published
Oct 1
Layer
Application layer
Setting
coding agent
Evaluated models
Meta-Llama-3.3-70B-Instruct, Meta-Llama-4-Scout-17B-16E-Instruct, Meta-Llama-3.1-8B-Instruct and others18
- AttacksMalicious use
Abstract作者称第三方外传可发生于正常网页获取，防护需考虑工具请求而不只考虑提示注入。
LLMLeak 是利用网页获取请求携带预先编码秘密的隐蔽通道攻击。它关注恶意软件已能读取数据、却因网络限制或监控无法直接外传的情形。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2609.35557Sep 29
论文：容器、权限规则与沙箱都无法区分读取研究代码的是编译器还是编码 Agent
评测容器、权限规则与沙箱能否区分编译器与编码 Agent 的文件读取
arXiv
2609.35557
Published
Sep 29
Layer
Application layer
Setting
coding agent
- DefensesPermissions and approval
- RiskDangerous agent actions
- ComponentPermissions and sandboxing
Abstract此 harness 表达不了编译器可读而智能体不可读，五条路径两边都不覆盖。
作者在物理流求解器上检查一套编码智能体配置：编译器必须读本组材料与反应模型，这些字节又不得进入模型上下文或离开本机的记录。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2609.32915Sep 27
AgentTell：浏览器 Agent 的行为侧信道泄漏基准
构建 AGENTTELL 基准，测量浏览器 Agent 的行为侧信道隐私泄露
arXiv
2609.32915
Published
Sep 27
Layer
Application layer
Setting
web agent
Evaluated models
Claude Sonnet 5, Gemini 3.7 Flash, GPT-5.6 Luna and others6
- RiskDangerous agent actions
Abstract论文评估了浏览器智能体的行为侧信道泄露风险，并基于近万次会话提供了评测数据。
Paper analysisRead analysis
OtherarXiv 2609.22720Sep 19
残障披露如何在对话式 AI 中流动：ChatGPT、Claude、Gemini 的记忆、隐私与语境完整性
访谈残障成人，分析向记忆型助手披露残障时的信息流与语境完整性
arXiv
2609.22720
Published
Sep 19
Layer
Application layer
Setting
Models and APIs
- ComponentMemory
Paper analysisRead analysis
DefensesarXiv 2609.19587Sep 17
红队测试 Auto Mode：用注入与多上下文攻击绕过编码 Agent 阻断监控器
提出 Auto Mode ++，加固编码 Agent 阻断监视器
arXiv
2609.19587
Published
Sep 17
Layer
Application layer
Setting
coding agent
Evaluated models
Opus 5, Opus 4.8, Opus 4.7 and others8
- DefensesMonitoring and auditing
- RiskDeception and scheming
- ComponentMonitors
Abstract系统评估生产阻断监视器在面对持续恶意智能体时的表现，提出加固防御并指明多上下文防御挑战。
Paper analysisRead analysis
AttacksarXiv 2609.18217Sep 16
研究：跨通道碎片化攻击可绕过 MCP 单通道防御
提出跨通道分段攻击，利用 MCP 隐式信任外发敏感文件
arXiv
2609.18217
Published
Sep 16
Layer
Application layer
Setting
coding agent
Evaluated models
GPT-4o, GPT-4o-mini, GPT-5.4 and others15
- AttacksPrompt injection
- TechniqueEncoding and obfuscation
- ComponentMCP and skills
Abstract论文揭示了 MCP 多通道隐式信任层级，通过跨通道分段与协议级利用突破现有防御，展示了严重的凭证外发风险。
Paper analysisRead analysis
AttacksarXiv 2609.13889Sep 12
PMPA：针对 OpenClaw 与 Claude Code 的持久记忆投毒攻击
提出 PMPA，经外部源向 Agent 持久记忆投毒并跨会话泄露隐私
arXiv
2609.13889
Published
Sep 12
Layer
Application layer
Setting
coding agent
Evaluated models
DeepSeek-V4-Flash, DeepSeek-V4-Pro, Qwen3-Max
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentMemory
Abstract论文提出了持久记忆投毒攻击 PMPA，报告了 harness 智能体在处理外部数据时易受跨会话记忆投毒与隐私泄露的风险。
Paper analysisRead analysis
AttacksarXiv 2609.12911Sep 11
研究：置换式保密在混合 FHE 推理中失效，可精确恢复 ResNet-20 全部线性层
攻破混合 FHE 推理的置换式保密，精确恢复线性层排序谱
arXiv
2609.12911
Published
Sep 11
Layer
Infrastructure layer
Setting
Models and APIs
Evaluated models
ResNet-20, ResNet-56, ResNet-50 and others12
- AttacksExtraction and theft
Paper analysisRead analysis
DefensesarXiv 2609.12378Sep 11
Odin：面向隐私保护 Llama 3 8B 推理的开源端到端 FHE 实现
提出端到端全同态加密方案 ODIN，在云端密文上完成推理以保护提示词
arXiv
2609.12378
Published
Sep 11
Layer
Infrastructure layer
Setting
Models and APIs
Evaluated models
Meta-Llama-3-8B
AbstractODIN 用可复用的特征主序打包和联合误差预算，在单张 H100 上完成 Llama-3-8B 的 CKKS 推理。
ODIN 是面向 Llama 的服务器侧 CKKS 推理系统：客户加密提示词，诚实但好奇的服务器用明文权重在密文上算完 Llama-3，只返回结果密文。
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 2608.25776Aug 26
EvoMal attack uses self-poisoning of self-evolving coding agents to achieve worm-like propagation
提出 EVOMAL，诱导自演化编码 Agent 在创建技能时自我投毒并蠕虫传播
arXiv
2608.25776
Published
Aug 26
Layer
Application layer
Setting
coding agent
Evaluated models
Devstral-Small-2, Gemma-4-31B-IT, Qwen3-Coder-Next and others7
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentMCP and skills
Abstract论文指出了自演化智能体模仿检索代码的自我毒化漏洞，提出 EVOMAL 攻击与防御，阐明了供应链新风险。
Paper analysisRead analysis
DefensesarXiv 2608.25431Aug 26
GIFT: Enforcing User Data Isolation in LLM Serving via GPU Information Flow Tracking
Propose GIFT, isolating user data in shared LLM serving using encryption and GPU information flow tracking
arXiv
2608.25431
Published
Aug 26
Layer
Infrastructure layer
Setting
Models and APIs
Evaluated models
Qwen2.5-14B, Qwen2.5-32B, Qwen2.5-72B and others6
- DefensesInstruction–data separation
- AttacksExtraction and theft
AbstractGIFT uses cryptographic isolation plus GPU information flow rules to enforce user data isolation on vLLM and DistServe, and reports throughput overhead.
GIFT is the authors' user data isolation system for shared LLM serving, with GIFT-CC enabling confidential computing. In the conclusion, the authors state it is the first IFT design for this scenario.
Paper analysisRead analysis
DefensesarXiv 2608.24017Aug 25
WebMCP-Phalanx：为浏览器内 LLM Agent 工具调用建立信任边界
提出 WebMCP-Phalanx，用能力凭证与双智能体隔离浏览器工具调用
arXiv
2608.24017
Published
Aug 25
Layer
Application layer
Setting
web agent
- DefensesPermissions and approval
- AttacksPrompt injection
- ComponentMCP and skills
AbstractPhalanx 用浏览器凭证绑定工具所有权，Q/P 分离拦住描述注入，工具名与返回值仍有残留。
WebMCP-Phalanx针对浏览器里的 WebMCP 智能体：在 SOP 下补工具所有权，并把语义检查从有工具权的执行中分开。
Paper analysisRead analysis
AttacksarXiv 2608.17485Aug 18
KeyPooling: Measuring where LLM API relay paths break prompt cache isolation
Proposing KeyPooling to measure where API relays break prompt cache isolation
arXiv
2608.17485
Published
Aug 18
Layer
Infrastructure layer
Setting
Models and APIs
Evaluated models
grok-4-fast-non-reasoning, grok-4-1-fast-non-reasoning, grok-3 and others11
- AttacksExtraction and theft
- TechniqueSide channels
AbstractShared upstream credentials place relay customers in the same cache domain; preserving identity or separate principals can separate them.
KeyPooling measures where LLM API relays drop customer identity, causing prompt caches to no longer be isolated per downstream customer.
Paper analysisRead analysis
AttacksarXiv 2608.09577Aug 10
ElasticBack：通过触发词与规则耦合在 LLM Agent 技能中植入条件后门
提出 ElasticBack，在单个 Agent 技能中植入条件后门
arXiv
2608.09577
Published
Aug 10
Layer
Application layer
Setting
coding agent
Evaluated models
GLM-5.2, MiniMax-M3, GPT-5.4 and others4
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentMCP and skills
AbstractElasticBack 用冻结规则加遗传搜索演化 trigger，在单 skill 上做条件后门。
ElasticBack 是放在 LLM agent 单个 skill 里的条件后门攻击。。
Paper analysisRead analysis

AI Safety Daily

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