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：Detection evasion Attack type：Malicious use System component：Monitors System component：Vision Clear all

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

DefensesarXiv 2610.09793Oct 7
用时序逻辑监控工具 Agent
提出 MFOTL 离线时序验证以监控工具 Agent 的危险动作链
arXiv
2610.09793
Published
Oct 7
Layer
Application layer
Setting
AI Agent
Attacker
Black-box
- DefensesMonitoring and auditing
- AttacksPrompt injection
- ComponentMonitors
Abstract形式化验证可有效识别多步攻击，但其有效性高度依赖轨迹的可信历史记录。
这篇论文对工具调用大语言模型智能体的安全监控进行了形式化时序验证研究。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2609.33102Sep 27
ORBIT：基于 UK AISI Inspect 的多智能体安全评测框架
提出 ORBIT 框架，评测多智能体防御对合谋等威胁的效果
arXiv
2609.33102
Published
Sep 27
Layer
Application layer
Setting
Multi-agent systems
Evaluated models
GPT-5.4, Claude Sonnet 4.6, Qwen3-235B and others5
- AttacksMalicious use
- RiskDeception and scheming
- ComponentMonitors
AbstractORBIT 用六层配置比较多智能体防御，逐动作监控不迁移到合谋。
Paper analysisRead analysis
Risky behaviorarXiv 2609.30266Sep 25
研究显示 Claude Code、Codex 等本地智能体可轻易篡改自身执行轨迹
测试编程智能体被诱导或自主篡改自身执行轨迹的行为
arXiv
2609.30266
Published
Sep 25
Layer
Application layer
Setting
coding agent
Evaluated models
GPT-5.6-Sol, GPT-6-Sol, Opus-5 and others11
- AttacksMalicious use
- RiskDeception and scheming
- ComponentMonitors
Abstract论文评估了本地智能体的轨迹防篡改能力，发现其可因外部指令或奖励追求而清除执行证据，亟需建立独立拦截记录机制。
这是一项针对本地大模型智能体执行轨迹完整性与安全防篡改边界的实证研究。
Paper analysisRead analysis
AttacksarXiv 2609.27542Sep 23
Researchers use control token injection to delete gpt-oss-20b's reasoning chain and bypass chain-of-thought monitoring
提出控制标记注入，清空工具 Agent 的思维链以绕过推理监控
arXiv
2609.27542
Published
Sep 23
Layer
Application layer
Setting
AI Agent
Evaluated models
openai/gpt-oss-20b, google/gemma-4-26B-A4B-it, deepseek-ai/DeepSeek-R1-Distill-Qwen-7B and others4
- AttacksPrompt injection
- TechniqueConversation structure manipulation
- ComponentReasoning chains
Abstract论文阐明控制标记注入可抑制 CoT 使监控失效并促成拒绝绕过，且工具执行受解析器宽容度影响。
Paper analysisRead analysis
DefensesarXiv 2609.22724Sep 19
MATE：面向移动 Agent 的策略感知安全审计方法
提出移动 Agent 轨迹审计器 MATE，判断轨迹是否违反可编辑安全策略
arXiv
2609.22724
Published
Sep 19
Layer
Application layer
Setting
Browser and computer use
Attacker
Black-box
Evaluated models
MATE-0.5B, MATE-1.5B, MATE-3B
- DefensesMonitoring and auditing
- AttacksMalicious use
- RiskDangerous agent actions
- ComponentMonitors
AbstractMATE 用合成轨迹微调轻量审计器，在可编辑策略下判断移动智能体轨迹是否违规并给出解释。
MATE 是面向移动 GUI 智能体的轻量策略条件轨迹审计器，在给定自然语言安全策略时判断轨迹是否违规并解释原因。
Paper analysisRead analysis
DefensesarXiv 2609.21363Sep 18
用扩散模型隐空间扰动防御视觉语言模型的地理位置隐私泄露
提出扩散潜空间地理扰动防御，抑制视觉语言模型泄露精确地理位置
arXiv
2609.21363
Published
Sep 18
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
GPT-4.1, GPT-5, Gemini 2.5 Pro and others5
- AttacksMalicious use
- ComponentVision
Paper analysisRead analysis
AttacksarXiv 2609.15989Sep 15
Researchers propose plan injection attack that can bypass chain-of-thought monitoring
提出 plan injection 攻击，植入良性伪装计划以绕过思维链监视器
arXiv
2609.15989
Published
Sep 15
Layer
Prompt layer
Setting
Models and APIs
Attacker
Gray-box
Evaluated models
Qwen2.5-Coder-7B-Instruct, Qwen3-8B, DeepSeek-R1-0528 and others6
- AttacksDetection evasion
- TechniqueReasoning-chain manipulation
- ComponentMonitors
Abstract论文指出模型会将外部上下文中的伪装计划改写为自身推理，使思维链监视器失效甚至反向合理化恶意行为。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2609.06966Sep 7
Mole：面向前沿实验室 AI 智能体的内部威胁检测基准
提出 MOLE 基准，评测监控器对 AI 智能体内部威胁的检出能力
arXiv
2609.06966
Published
Sep 7
Layer
Application layer
Setting
AI Agent
Evaluated models
GPT-5.3 Chat, GPT-5.5, GPT-4.1 and others15
- DefensesMonitoring and auditing
- AttacksMalicious use
- ComponentMonitors
AbstractMOLE 构建了智能体内部威胁检测基准，作者指出高漏检现状与推理痕迹的防御价值，并展示了监控策略优化空间。
MOLE 是首个针对前沿 AI 实验室智能体内部威胁检测的开放基准，旨在评估在有限每日审查预算与日常研发业务背景下识别恶意账号的能力。。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2609.05843Sep 5
SinoGlyphBench：中文字形混淆下语言模型审核的诊断基准
提出 SinoGlyphBench，用配对字形混淆诊断中文审核稳定性
arXiv
2609.05843
Published
Sep 5
Layer
Prompt layer
Setting
Models and APIs
Evaluated models
GPT-5.4 mini, GPT-5.5, DeepSeek V4 Flash and others12
- AttacksDetection evasion
- TechniqueEncoding and obfuscation
- ComponentGuardrails and judges
AbstractSinoGlyphBench 显示中文字形混淆会同时增加有害漏报和误报，全文扰动最重。
SinoGlyphBench是一个配对诊断基准，用来看内容审核在可还原的中文字形混淆下是否还保持原决定。
Paper analysisRead analysis
DefensesarXiv 2608.12876Aug 13
SPARED：用对抗编辑数据训练基于推理的 AI 生成图像检测器
提出 SPARED，用对抗编辑数据训练基于推理的 AI 生成图像检测器
arXiv
2608.12876
Published
Aug 13
Layer
Training and data layer
Setting
Models and APIs
Attacker
Black-box, White-box
Evaluated models
Qwen3.5-9B
- DefensesDetection and filtering
- AttacksDetection evasion
- ComponentGuardrails and judges
AbstractSPARED 用防捷径的编辑对抗环训练 9B 推理检测器。
SPARED 是一套用于可解释 AI 生成图像检测的攻防强化学习流程，报告模型是 Qwen3.5-9B 上的第三轮防御器。
Paper analysisRead analysis
DefensesarXiv 2608.00716Aug 1
生成图像更易被遗忘：面向合成图像检测的机器遗忘视角
提出机器遗忘式生成图像检测，用遗忘前后特征相似度区分自然图与生成图
arXiv
2608.00716
Published
Aug 1
Layer
Model layer
Setting
Models and APIs
Evaluated models
DINOv2 ViT-S/14, DINOv2 ViT-B/14, DINOv2 ViT-L/14 and others9
- DefensesDetection and filtering
- AttacksMalicious use
- ComponentVision
Abstract作者用剪枝或 LoRA 遗忘拉开生成图与自然图的特征相似度，并在多个检测基准上报告结果。
这篇工作把 AI 生成图像检测写成对预训练视觉模型的机器遗忘：原模型对两类图像损失都低，剪枝后生成图特征变得更不像原模型。
Paper analysisRead analysis
DefensesarXiv 2607.14737Jul 16
GeoDetect：面向视觉语言预训练模型的几何对抗检测方法
提出 GeoDetect 以几何分数检测视觉语言预训练模型的对抗样本
arXiv
2607.14737
Published
Jul 16
Layer
Model layer
Setting
Models and APIs
Attacker
White-box
Evaluated models
CLIPViT (ViT-B/16), CLIPCNN (ResNet-50), ALBEF and others4
- DefensesDetection and filtering
- AttacksDetection evasion
- ComponentEmbeddings
AbstractGeoDetect 用几何分数检测 VLP 对抗样本。
GeoDetect 是面向视觉-语言预训练模型的对抗样本检测方法，用嵌入几何而不是任务 logits 区分干净样本与对抗样本。
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 2605.30040May 28
研究揭示 LLM 按 token 计费可被服务商系统性虚报
构造可通过计费审计的 token 虚报，使服务商系统性超收
arXiv
2605.30040
Published
May 28
Layer
Infrastructure layer
Setting
Models and APIs
Evaluated models
PALACE domain-adapted auditor
- AttacksDetection evasion
- TechniqueDefense-targeted
- ComponentMonitors
Abstract不诚实提供方可在三种 token 审计上虚报并通过检查，作者认为需改用提供方不控制的证据。
不透明 LLM 的按 token 计费里，现有审计核验的是提供方可控的证据，不诚实提供方可以虚报并通过检查。。
Paper analysisRead analysis
AttacksarXiv 2605.21541May 20
研究者提出 FRA-Attack 频域正则化方法，提升对闭源 MLLM 的迁移攻击效果
提出 FRA-Attack，用频域对齐与梯度正则做可迁移定向视觉攻击
arXiv
2605.21541
Published
May 20
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
GPT-5.2, GPT-5.4, Claude-Opus-4.6 and others15
- AttacksDetection evasion
- TechniqueGradient and representation optimization
- ComponentVision
AbstractFRA-Attack 以频域对齐和梯度低通做可迁移定向攻击，闭源面板上作者称整体更强，Gemma 除外。
FRA-Attack 是面向闭源 MLLM 的迁移式定向图像攻击。作者要解决的是：空间域局部对齐重复全局低频，代理梯度的高频成分又把更新拉向代理自身。
Paper analysisRead analysis
DefensesarXiv 2604.11806Apr 14
Meerkat：跨大量 Agent 轨迹检测安全违规
提出 Meerkat，聚类并搜索 Agent 轨迹库以定位跨轨迹安全违规
arXiv
2604.11806
Published
Apr 14
Layer
Application layer
Setting
AI Agent
Evaluated models
GPT-5.4, GPT-5.4-mini, Qwen3.5-397B-A17B and others5
- DefensesMonitoring and auditing
- AttacksMalicious use
- RiskReward hacking
- ComponentMonitors
AbstractMeerkat 用聚类与智能体搜索定位跨轨迹违规，并在真实评测轨迹中找出脚手架泄题与奖励黑客。
Meerkat 是事后审计方法，用来在智能体轨迹仓库里找出单条看不出、合在一起才违反自然语言安全属性的见证。
Paper analysisRead analysis
AttacksarXiv 2604.03117Apr 3
Unveiling Physical-World Semantic Vulnerabilities: Universal Adversarial Patches for Infrared Vision-Language Models
Proposes the universal curved mesh patch UCGP, causing cross-task semantic bias in infrared vision-language models.
arXiv
2604.03117
Published
Apr 3
Layer
Prompt layer
Setting
Models and APIs
Attacker
Gray-box
Evaluated models
OpenAI CLIP ViT-L/14, OpenCLIP ViT-B/16, Meta-CLIP ViT-L/14 and others10
- AttacksDetection evasion
- TechniqueGradient and representation optimization
- ComponentVision
AbstractUCGP uses one shared curved-grid patch to simultaneously lower the classification and open-ended outputs of IR-VLM.
UCGP is a universal physical adversarial patch for infrared vision-language models. The question is: can a patch that can be attached and reused across samples cause stable semantic deviations in the infrared visual evidence shared by classification, description, and VQA. The patch is constrained to a low-frequency curved mesh.
Paper analysisRead analysis
DefensesarXiv 2602.18252Feb 20
Adversarial robustness of discrete image tokenizers
提出无监督对抗微调，加固离散图像 tokenizer
arXiv
2602.18252
Published
Feb 20
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
UniTok, UniTok-MLLM, TiTok-BL128 and others7
- DefensesModel hardening
- AttacksDetection evasion
- ComponentVision
Abstract无监督嵌入攻击暴露离散 tokenizer 的弱点，只微调 encoder 即可提升多任务鲁棒准确率。
作者研究离散图像 tokenizer 的对抗鲁棒性，给出与任务无关的嵌入攻击，以及只微调 encoder 的无监督对抗训练。
Paper analysisRead analysis
DefensesarXiv 2602.12461Feb 13
Semantic-aware adversarial fine-tuning for CLIP
Propose SAFT, which uses semantic description ensembles to construct adversarial examples and fine-tune the image encoder
arXiv
2602.12461
Published
Feb 13
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
ViT-B/32, CLIP-L/14
- DefensesModel hardening
- AttacksDetection evasion
- ComponentVision
AbstractSAFT fine-tunes the CLIP image encoder with ensemble adversarial examples from refined class descriptions to improve the trade-off between zero-shot accuracy and robustness.
SAFT is a method for adversarially fine-tuning the CLIP image encoder to improve the accuracy–robustness trade-off in zero-shot classification.
Paper analysisRead analysis
AttacksarXiv 2602.09431Feb 10
Grounding-driven attack: Improving the transferability of encoder-based adversarial attacks to large vision-language models
Propose GDA, which aligns image adversarial perturbations with text-grounded evidence and transfers attacks to vision-language models
arXiv
2602.09431
Published
Feb 10
Layer
Prompt layer
Setting
Models and APIs
Attacker
No-box (offline transfer)
Evaluated models
LLaVA, Qwen2.5-VL, InternVL3 and others10
- AttacksDetection evasion
- TechniqueGradient and representation optimization
- ComponentVision
AbstractGDA aligns perturbations with text-grounded evidence, and the authors say this improves zero-query black-box transferability on heterogeneous LVLMs.
GDA is a zero-query black-box encoder transfer attack for heterogeneous LVLMs.
Paper analysisRead analysis

AI Safety Daily

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