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 technique：Multi-turn escalation Attack technique：Encoding and obfuscation Attack technique：Conversation structure manipulation System component：Reasoning chains System component：Vision Clear all

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

AttacksarXiv 2610.12644Oct 9
Cancerous backdoor attack: progressive activation via invisible tokens and a stage controller.
Proposes the Cancerous backdoor, which progressively corrupts reasoning using invisible stage tokens.
arXiv
2610.12644
Published
Oct 9
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Llama-3.1-8B-Instruct, Qwen2.5-7B-Instruct, Mistral-7B-Instruct-v0.3
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentReasoning chains
AbstractHidden tokens plus a stage controller make the backdoor progressively corrupt reasoning upon triggering, and the authors believe that looking only at the answer will miss it.
Cancerous Backdoor Attack changes LLM backdoors from one-time activation to multi-stage reasoning corruption that accumulates progressively with triggers.
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2609.35912Sep 28
MMSkillRisk：多模态技能图像中的隐藏指令可诱导 Agent 执行未授权操作
提出 MMSkillRisk 基准，评测多模态技能图像中的隐藏指令诱导 Agent 越权操作
arXiv
2609.35912
Published
Sep 28
Layer
Application layer
Setting
coding agent
Evaluated models
GPT-5.6-sol, Kimi-K2.6, Kimi-K3 and others6
- AttacksPrompt injection
- TechniqueCross-modal payloads
- ComponentMCP and skills
Abstract论文提出 MMSkillRisk 基准与 NCVA 攻击，揭示多模态技能中教学图像潜藏指令可引发未授权操作并与良性任务共存。
Paper analysisRead analysis
AttacksarXiv 2609.29775Sep 24
输出前缀攻击瞄准推理模型
提出推理通道预填与输出前缀组合攻击以越狱推理模型
arXiv
2609.29775
Published
Sep 24
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
Gemini 3 Flash Preview, DeepSeek V4 Flash, Claude Haiku 4.5
- AttacksJailbreaks
- TechniqueReasoning-chain manipulation
- ComponentReasoning chains
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
AttacksarXiv 2609.09553Sep 9
研究：任意密文攻击前沿大模型无需微调即可越狱
提出任意字母置换密码越狱，无需微调即可诱导有害输出
arXiv
2609.09553
Published
Sep 9
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
Claude Sonnet 4, Claude Sonnet 4.5, Gemini 3 Flash (preview) and others7
- AttacksJailbreaks
- TechniqueEncoding and obfuscation
- ComponentGuardrails and judges
Abstract前沿大模型无需微调即可通过提示词学会字母置换密码，进而绕过分类器与对齐，作者呼吁关注能力提升带来的安全风险。
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
AttacksarXiv 2608.09867Aug 11
Researchers steal proprietary LLM inference chains via cross-model compatibility of encrypted inference blocks
利用加密思维块的跨模型重放，诱导弱模型逐字转写专有推理链
arXiv
2608.09867
Published
Aug 11
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
Claude Opus 4.8, Claude Sonnet 4.6, Claude Haiku 4.5 and others13
- AttacksExtraction and theft
- TechniqueReasoning-chain manipulation
- ComponentReasoning chains
Abstract论文揭示了客户端加密思维块跨模型互换漏洞，提出了低成本提取方法并给出防御方案。
Paper analysisRead analysis
AttacksarXiv 2608.04034Aug 3
HACA：原子越狱策略解耦组合的多模态自动红队方法
提出 HACA，将图文越狱拆为原子策略并自动组合以诱导有害输出
arXiv
2608.04034
Published
Aug 3
Layer
Prompt layer
Setting
Models and APIs
Evaluated models
LLaVA-1.6-Mistral-7B, InternVL-3.5-8B, VLGuard (LLaVA-v1.5-7B-Mixed) and others5
- AttacksJailbreaks
- TechniqueCross-modal payloads
- ComponentVision
Paper analysisRead analysis
AttacksarXiv 2606.01837Jun 1
论文提出 DSR 跨模态越狱框架：用良性输入诱导 MLLM 有害输出
提出 DSR 跨模态越狱，用无害图文诱导 MLLM 生成有害图像
arXiv
2606.01837
Published
Jun 1
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
gemini-2.5-flash-image, qwen-image-2.0, Tongyi Wanx 2.5
- AttacksJailbreaks
- TechniqueCross-modal payloads
- ComponentImage generation
AbstractDSR 用无害图文诱导商业 MLLM 输出有害图像，作者称组合能力会反过来扩大攻击面。
DSR 是一种 black-box 跨模态越狱：输入侧不放显式有害内容，而让商业 MLLM 在推理中把无害图文重组成有害图像。
Paper analysisRead analysis
AttacksarXiv 2605.18915May 18
DMN：面向多图输入的多模态大模型组合式越狱框架
提出多图越狱框架 DMN，把有害意图拆进图像序列绕过安全护栏
arXiv
2605.18915
Published
May 18
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
GPT-4V, GPT-4o, GPT-5 and others10
- AttacksJailbreaks
- TechniqueCross-modal payloads
- ComponentVision
Paper analysisRead analysis
AttacksarXiv 2605.05709May 7
研究者提出利用重构与隐藏权衡的多模态大模型越狱方法
提出删字变体越狱方法，利用重构与隐藏权衡绕过多模态模型安全过滤
arXiv
2605.05709
Published
May 7
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
GPT-5.4-nano, GPT-5.4-mini, Gemini-2.5-Flash and others19
- AttacksJailbreaks
- TechniqueEncoding and obfuscation
- ComponentVision
Paper analysisRead analysis
AttacksarXiv 2605.00583May 1
Jailbreaking vision-language models through the visual modality
提出四类视觉越狱，把有害意图编码进图像以绕过安全拒答
arXiv
2605.00583
Published
May 1
Layer
Prompt layer
Setting
Models and APIs
Evaluated models
GPT-5.2, Claude-Haiku-4.5, Gemini-3-Flash and others8
- AttacksJailbreaks
- TechniqueCross-modal payloads
- ComponentVision
Abstract四类视觉越狱在六个 VLM 上检验文本对齐能否挡住视觉意图，模型与攻击之间差异大。
四类视觉越狱用来检验前沿 VLM 的文本安全训练能否挡住由图像传达的有害意图。。
Paper analysisRead analysis
AttacksarXiv 2603.10091Mar 10
Multi-Stream Perturbation Attack: Breaking the Safety Alignment of Reasoning LLMs through Concurrent Task Interference
提出多流扰动攻击，交织有害与良性任务绕过思考型 LLM 安全对齐
arXiv
2603.10091
Published
Mar 10
Layer
Prompt layer
Setting
Models and APIs
Evaluated models
Qwen3 1.7B, Qwen3 4B, Qwen3 8B and others11
- AttacksJailbreaks
- TechniqueEncoding and obfuscation
- ComponentReasoning chains
Abstract多流扰动用任务交织测试 thinking 模式，作者称它同时冲击内容安全与推理稳定性。
多流扰动攻击把有害任务与良性任务织进同一提示词，测试 thinking 模式的内容安全与推理稳定性。作者认为，交织多任务会拉长思考，使安全检测面对被打散的有害意图，并可能把推理用于合理化，或陷入崩溃与重复。方法不访问梯度。
Paper analysisRead analysis
AttacksarXiv 2603.07590Mar 8
Model as Lego builder: assembling malicious content from harmless blocks using a semantic blueprint
Propose StructAttack, which uses structural diagrams to induce visual models to assemble malicious content from harmless semantic slots.
arXiv
2603.07590
Published
Mar 8
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
GPT-4o (1120), Gemini-2.0-Flash (001), Gemini-2.5-Flash and others6
- AttacksJailbreaks
- TechniqueCross-modal payloads
- ComponentVision
AbstractStructAttack decomposes a harmful query into harmless slots and draws them into a structure graph, inducing the LVLM to fill in harmful content under a black-box single query.
StructAttack is a black-box, single-query jailbreak against LVLMs: it decomposes a harmful instruction into separate seemingly harmless slots, draws them into a structural diagram, and then makes the model reassemble the slots into harmful content.
Paper analysisRead analysis
AttacksarXiv 2603.00565Feb 28
MIDAS: Multi-Image Dispersion and Semantic Reconstruction for MLLM Jailbreaking
提出 MIDAS ，把有害语义拆进多张视觉谜题并跨图重构以越狱多模态模型
arXiv
2603.00565
Published
Feb 28
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
GPT-4o, GPT-5-Chat, Gemini-2.5-Pro and others7
- AttacksJailbreaks
- TechniqueCross-modal payloads
- ComponentVision
Paper analysisRead analysis
AttacksarXiv 2602.18514Feb 20
招聘中的特洛伊木马：标准模型与推理模型间接提示注入的红队测试案例研究
用简历中的不可见指令做间接提示注入，操纵招聘推荐
arXiv
2602.18514
Published
Feb 20
Layer
Application layer
Setting
LLM applications
Evaluated models
Qwen 3 30B A3B Instruct 2507, Qwen 3 30B A3B Thinking 2507
- AttacksPrompt injection
- TechniqueEncoding and obfuscation
- ComponentReasoning chains
Abstract作者称推理在简单注入下会帮攻击圆谎，在不合逻辑的复杂注入下又会把欺骗漏出来。
Paper analysisRead analysis
AttacksarXiv 2603.10010Feb 18
FERRET：基于扩展的多模态自动化红队测试框架
提出 FERRET 框架，用扩展生成多轮多模态越狱对话
arXiv
2603.10010
Published
Feb 18
Layer
Prompt layer
Setting
Models and APIs
Evaluated models
Llama Maverick, Claude Haiku, GPT-4o
- AttacksJailbreaks
- TechniqueAutomated search
- ComponentVision
Paper analysisRead analysis
AttacksarXiv 2602.14399Feb 16
Multi-turn Adaptive Prompt Attack against Large Vision-Language Models
Propose MAPA multi-turn adaptive prompt attack, alternating text and visual actions to jailbreak vision-language models
arXiv
2602.14399
Published
Feb 16
Layer
Prompt layer
Setting
Models and APIs
Evaluated models
LLaVA-V1.6-Mistral-7B, Qwen2.5-VL-7B-Instruct, Llama-3.2-Vision-11B-Instruct and others4
- AttacksJailbreaks
- TechniqueMulti-turn escalation
- ComponentVision
Paper analysisRead analysis
AttacksarXiv 2602.10148Feb 10
Red-Teaming Multimodal Reasoning: Cross-Modal Entanglement Attacks Jailbreak Vision-Language Models
Propose COMET cross-modal entanglement attack to jailbreak vision-language models
arXiv
2602.10148
Published
Feb 10
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
GPT-4.1, GPT-4.1-mini, Gemini-2.5-Flash and others9
- AttacksJailbreaks
- TechniqueCross-modal payloads
- ComponentVision
AbstractCOMET performs black-box cross-modal entanglement jailbreak.
COMET is a black-box jailbreak for VLMs with multimodal reasoning, used to red-team cross-modal safety alignment. The gap the authors address is: distributing malicious cues into fixed combinations of image and text cannot detect vulnerabilities that only appear in multi-step reasoning. The attack does not access gradients; it only constructs image and text inputs.
Paper analysisRead analysis
AttacksarXiv 2603.20198Feb 5
研究者提出 MM-Plan 多模态越狱框架
提出 MM-Plan 多模态越狱框架，自动规划多轮图像编辑以越狱视觉独占目标
arXiv
2603.20198
Published
Feb 5
Layer
Prompt layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
Llama-3.2-11B, InternVL3-8B, Qwen3-VL-8B and others8
- AttacksJailbreaks
- TechniqueAutomated search
- ComponentVision
Abstract作者称全局规划能攻破对单轮攻击更稳的前沿 MLLM。
MM-Plan 是面向 Visual Exclusivity 的 black-box 多模态红队方法，并配有人工基准 VE-Safety。
Paper analysisRead analysis

AI Safety Daily

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