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

Oct 13 · Tuesday

All papers

Found 19 papers

Filter5

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.

System type：LLM applications System type：coding agent Attack surface / entry point：Tool responses Attack surface / entry point：Memory writes Attack surface / entry point：Training data Clear all

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

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
AttacksarXiv 2610.10000Oct 7
超越奖励压制：有界奖励热启动 Bandit 的近最优离线攻击
提出热启动 Bandit 的部署前有界离线注入攻击
arXiv
2610.10000
Published
Oct 7
Layer
Training and data layer
Setting
LLM applications
Evaluated models
UCB, Thompson Sampling, ϵ-greedy
- AttacksPoisoning and backdoors
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.07639Oct 6
HarnessSecurity-Bench horizontally evaluates the security mechanisms of six coding agent frameworks.
评测编码智能体 harness 安全机制在对抗任务中的防护效果
arXiv
2610.07639
Published
Oct 6
Layer
Application layer
Setting
coding agent
- AttacksPrompt injection
- ComponentPermissions and sandboxing
Paper analysisRead analysis
AttacksarXiv 2610.05793Oct 5
研究者训练出按多智能体拓扑触发的代码后门模型
提出拓扑条件后门，使代码模型在推断多智能体部署时植入漏洞
arXiv
2610.05793
Published
Oct 5
Layer
Training and data layer
Setting
coding agent
Evaluated models
Qwen2.5-7B-Instruct, Qwen2.5-Coder-7B-Instruct
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentFine-tuning and open weights
Abstract论文证实了拓扑条件后门的可训练性与泛化性，展示了常规评测盲区，并提出差异化拓扑审计作为检测防御方向。
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 2609.35659Sep 29
Tracekit: A tamper-proof intent-reasoning-action audit system for autonomous coding agents
提出 Tracekit，审计编码 Agent 的意图、推理与动作并检测账本篡改
arXiv
2609.35659
Published
Sep 29
Layer
Application layer
Setting
coding agent
Evaluated models
Claude Code (claude -p, version 2.1), 独立评审（同一 CLI，无工具）
- DefensesMonitoring and auditing
- AttacksPrompt injection
- RiskDangerous agent actions
- ComponentMonitors
AbstractTracekit 把意图、自述和动作写入可锚定的哈希链。
Tracekit 是无第三方依赖的编码智能体审计层：把人类请求、模型自述和已执行动作并排写入可外部锚定的哈希链，执行前用正则门，事后用规则和独立模型互查。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2609.32616Sep 26
Cave-Bench：虚假指控下 LLM 智能体最高 60.06% 的运行会破坏正确工作
提出 CAVE-BENCH，评测智能体在虚假指责下破坏已完成的正确工作
arXiv
2609.32616
Published
Sep 26
Layer
Application layer
Setting
coding agent
Evaluated models
Claude-Sonnet-5, Claude-Opus-5, GPT-5.6-Sol and others14
- AttacksPrompt injection
- RiskDangerous agent actions
Abstract论文提出了评估智能体在虚假指责下破坏已有正确成果的 CAVE-BENCH，发现强模型易推翻已知证据，提出了门控缓解方案。
Paper analysisRead analysis
Analysis and theoryarXiv 2609.23215Sep 20
研究审计 Active Inference Agent 的 LLM 解释器失败模式
审计 Active Inference Agent 上 LLM 解释器的监督叙述失败
arXiv
2609.23215
Published
Sep 20
Layer
Application layer
Setting
LLM applications
Evaluated models
GPT-4o, Claude-3-Opus, Gemini
- AttacksPrompt injection
- RiskMisalignment and value divergence
Abstract三组触发下解释流畅但错误，所提缓解都未被评估。
作者审计的是挂在智能体上的运行时解释器，而不是智能体本身。操作员读到的是信念与动作的自然语言叙述。作者要问哪些可复现触发会让这段叙述失败，以及失败痕迹是什么样。
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 2609.17817Sep 16
论文：用投毒基准污染自改进编码 Agent，令其后续版本写出漏洞代码
提出基准投毒攻击，使自修改编码 Agent 跨代写出含漏洞代码
arXiv
2609.17817
Published
Sep 16
Layer
Application layer
Setting
coding agent
Evaluated models
Claude Sonnet 4.5, Qwen3.5-397B, gpt-oss-120b and others5
- AttacksPoisoning and backdoors
Abstract论文报告了恶意基准能污染自修改代码智能体的进化过程并持续输出漏洞，警示系统设计需纳入安全约束。
Paper analysisRead analysis
DefensesarXiv 2609.10707Sep 10
面向 AI-SOC 的神经符号防御框架：用 SIEM 解码器与 NeMo Guardrails 阻断日志投毒提示注入
提出神经符号纵深架构，阻断 AI-SOC 的日志投毒与间接提示注入
arXiv
2609.10707
Published
Sep 10
Layer
Application layer
Setting
LLM applications
Evaluated models
Gemma 3 (12B)
- DefensesDetection and filtering
- AttacksPrompt injection
- ComponentGuardrails and judges
Paper analysisRead analysis
AttacksarXiv 2609.01222Sep 1
Researchers propose context privilege escalation attack, affecting 12 agent frameworks
提出针对 Agent Harness 的上下文特权提升攻击
arXiv
2609.01222
Published
Sep 1
Layer
Application layer
Setting
coding agent
Evaluated models
GPT-5.5, GPT-5.4 mini, Claude-Sonnet-4.6 and others7
- AttacksPrompt injection
- TechniqueDormant triggers
- ComponentMemory
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2608.12851Aug 13
The researchers propose SkillMisevo-Bench to quantify the risk of skill mis-evolution in self-improving agents.
提出 SkillMisevo-Bench，量化自改进 Agent 的技能误演化
arXiv
2608.12851
Published
Aug 13
Layer
Application layer
Setting
coding agent
Evaluated models
Claude Code, Codex, Hermes and others4
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentMemory
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2607.14611Jul 16
论文评测 Claude Code 与 Codex 记忆文件中的提示注入风险
评测编码 Agent 记忆文件中提示注入的跨会话影响
arXiv
2607.14611
Published
Jul 16
Layer
Application layer
Setting
coding agent
Evaluated models
Claude Haiku 4.5, Claude Opus 4.7, GPT-5.2 and others4
- AttacksPoisoning and backdoors
- ComponentMemory
Abstract已植入记忆文件的载荷可以跨会话影响 Claude Code 与 Codex，攻击成功与载荷是否留下是两种不同的失败。
Bad Memory 在沙箱合成工作区里评测文件型持久记忆上的提示注入，比较载荷能否改变当前会话，以及能否在后续会话中留下或叠加。
Paper analysisRead analysis
AttacksarXiv 2607.11698Jul 13
AHA 用自动研究循环发现生产 Agent 漏洞概念
提出 AHA 自动科研循环，发现生产 Agent 的可复用漏洞概念
arXiv
2607.11698
Published
Jul 13
Layer
Application layer
Setting
coding agent
Attacker
Black-box
Evaluated models
Minimax-M2.7, Kimi-K2.6, Deepseek-V4-Pro and others5
- AttacksPrompt injection
- TechniqueAutomated search
Abstract论文提出基于可证伪科研循环的 AHA 框架，从生产级智能体中挖掘并沉淀漏洞概念图，提升攻击复用率并指导防御加固。
AHA 是一项针对生产级 LLM 智能体自动化红队测试的研究，旨在解决传统方法仅保留具体攻击载荷而缺乏成因解释、导致跨版本复用困难且无法指导修复的问题。
Paper analysisRead analysis
DefensesarXiv 2606.22673Jun 22
Agent MechSuits：面向多轮 CLI Agent 的机制子空间安全引导
提出 Agent MechSuits，对多轮 CLI Agent 做步级检测与子空间转向
arXiv
2606.22673
Published
Jun 22
Layer
Application layer
Setting
coding agent
Evaluated models
LLaMA-3.1-8B, Qwen-2.5-7B, Gemma-2-9B
- DefensesInference-time intervention
- AttacksMalicious use
- ComponentMonitors
Abstract白盒步级探针加稀疏转向可压低三模型恶意动作，同时提高生成崩溃率。
Agent MechSuits 是面向多轮 CLI 智能体的白盒、步级表示防御，并配有可执行轨迹基准 MAS。
Paper analysisRead analysis
AttacksarXiv 2604.03890Apr 5
From prompts to physical actions: structured backdoor attacks against LLM-mediated robot control systems
Propose a structured JSON backdoor that triggers malicious actions in robot control via a poisoned LoRA
arXiv
2604.03890
Published
Apr 5
Layer
Training and data layer
Setting
Embodied systems and robotics, LLM applications
Evaluated models
Llama-3.1-8B-Instruct, Gemma-2-9B-IT, DeepSeek-R1-Distill-Llama-8B and others4
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentFine-tuning and open weights
Abstract推理层后门传不到执行。
Paper analysisRead analysis
AttacksarXiv 2602.15344Feb 17
ER-MIA: Black-box Adversarial Memory Injection Attacks against Long-term Memory-augmented Large Language Models
提出 ER-MIA 黑盒记忆注入，用嵌入相近记忆抢占检索并改错后续回答
arXiv
2602.15344
Published
Feb 17
Layer
Application layer
Setting
LLM applications
Evaluated models
Llama3.2:3b, Gemma 3:27b, GPT-oss:20b
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentMemory
Paper analysisRead analysis
AttacksarXiv 2602.06409Feb 6
VENOMREC: Cross-modal Interactive Poisoning for Targeted Promotion in Multimodal LLM Recommendation Systems
提出 VenomRec 跨模态交互投毒，定向推广多模态推荐物品
arXiv
2602.06409
Published
Feb 6
Layer
Training and data layer
Setting
LLM applications
Evaluated models
VIP5 (T5-small), VIP5 (T5-base), DiffMM
- AttacksPoisoning and backdoors
- TechniqueCross-modal payloads
- ComponentVision
Paper analysisRead analysis

You have reached the end

AI Safety Daily

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

All papers

Attack type

Attack technique

Defense type

Impact

System type

Attack surface / entry point

Layer

System component

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

超越奖励压制：有界奖励热启动 Bandit 的近最优离线攻击

HarnessSecurity-Bench horizontally evaluates the security mechanisms of six coding agent frameworks.

研究者训练出按多智能体拓扑触发的代码后门模型

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

Tracekit: A tamper-proof intent-reasoning-action audit system for autonomous coding agents

Cave-Bench：虚假指控下 LLM 智能体最高 60.06% 的运行会破坏正确工作

研究审计 Active Inference Agent 的 LLM 解释器失败模式

研究：跨通道碎片化攻击可绕过 MCP 单通道防御

论文：用投毒基准污染自改进编码 Agent，令其后续版本写出漏洞代码

面向 AI-SOC 的神经符号防御框架：用 SIEM 解码器与 NeMo Guardrails 阻断日志投毒提示注入

Researchers propose context privilege escalation attack, affecting 12 agent frameworks

The researchers propose SkillMisevo-Bench to quantify the risk of skill mis-evolution in self-improving agents.

论文评测 Claude Code 与 Codex 记忆文件中的提示注入风险

AHA 用自动研究循环发现生产 Agent 漏洞概念

Agent MechSuits：面向多轮 CLI Agent 的机制子空间安全引导

From prompts to physical actions: structured backdoor attacks against LLM-mediated robot control systems

ER-MIA: Black-box Adversarial Memory Injection Attacks against Long-term Memory-augmented Large Language Models

VENOMREC: Cross-modal Interactive Poisoning for Targeted Promotion in Multimodal LLM Recommendation Systems