Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 233papers have no tags yet and are not included in filtered results.

Attack type：Poisoning and backdoors Attack type：Model tampering Attack surface / entry point：Weights and gradients Layer：Model layer Clear all

Another 233papers do not have category tags yet and are not included in filtered results.

AttacksarXiv 2610.13928Oct 10
AI models can provably hide arbitrary capabilities
Proposes digital lock compilation, encrypting executable networks into host weights to hide arbitrary capabilities
arXiv
2610.13928
Published
Oct 10
Layer
Model layer
Setting
Models and APIs
Evaluated models
Gemma 3 270M
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
AbstractThe authors use a digital lock to hide an executable capability and state that elicitation on the chemistry and parity tasks did not recover the capability.
The authors construct a class of backdoors that encrypt an executable network into the weights, used to discuss the limits of white-box elicitation audits under supply chain attacks.
Paper analysisRead analysis
InterpretabilityarXiv 2610.09000Oct 7
研究定位 LLaMA-2-7B-Chat 安全敏感参数
定位安全敏感参数并稀疏改动权重以削弱安全对齐
arXiv
2610.09000
Published
Oct 7
Layer
Model layer
Setting
Models and APIs
Evaluated models
LLaMA-2-7B-Chat
- AttacksModel tampering
- RiskInappropriate refusal
Paper analysisRead analysis
AttacksarXiv 2610.06049Oct 5
研究者提出仅修改 SAE 解码器的后门攻击，可在不改动语言模型的情况下植入行为
提出只改 SAE 解码器的后门，在冻结语言模型上植入代码插入
arXiv
2610.06049
Published
Oct 5
Layer
Model layer
Setting
Models and APIs
Evaluated models
Gemma 3 4B Instruct, Qwen3 1.7B, Llama 3.1 8B Instruct
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
Abstract只改解码器的 SAE 可带上代码插入后门，且若干常规质量指标变化不大。
只改 SAE 解码器、冻结编码器与语言模型，即可把代码插入行为放进推理时的辅助组件。作者把这视为 SAE 供应链问题：checkpoint 一旦替换某一层激活，就参与后续计算，而不只是解释工具。
Paper analysisRead analysis
AttacksarXiv 2610.05894Oct 5
研究者提出闭环偏见注入攻击
提出闭环激活引导，在扩散语言模型去噪过程中注入定向偏见
arXiv
2610.05894
Published
Oct 5
Layer
Model layer
Setting
Models and APIs
Attacker
Gray-box
Evaluated models
LLaDA-8B-Instruct, Dream-v0-Instruct-7B, LLaDA-MoE-7B-A1B-Instruct
- AttacksModel tampering
- TechniqueGradient and representation optimization
Abstract通过去噪中间概率反馈动态调节残差流转向，证明扩散语言模型在推理服务栈中存在闭环偏见注入脆弱性。
Paper analysisRead analysis
AttacksarXiv 2610.03124Oct 2
研究提出 Untag 攻击框架，使开源大模型触发标签滥用检测机制失效
提出 UNTAG，绕过开源权重模型的 trigger-tag 滥用检测
arXiv
2610.03124
Published
Oct 2
Layer
Model layer
Setting
Models and APIs
Attacker
Black-box, White-box
Evaluated models
Llama2 (7B), Llama3 (8B), Qwen2.5 (7B)
- AttacksDetection evasion
- TechniqueDefense-targeted
- ComponentFine-tuning and open weights
Abstract作者用 UNTAG 评估 trigger-tag，认为其不能充当开放权重滥用检测的主要边界。
UNTAG 是针对开放权重 LLM 上条件滥用检测 trigger-tag 的统一攻击评估：信号可以在解码期注入，也可以写进权重，但攻击者若能改输出或改已发布权重，检测可以失效。
Paper analysisRead analysis
AttacksarXiv 2610.01062Oct 1
Kernelized Activation Steering：将激活引导推广到再生核希尔伯特空间的新框架
提出核化激活引导 KAS，在推理时自适应转向激活以诱导越狱
arXiv
2610.01062
Published
Oct 1
Layer
Model layer
Setting
Models and APIs
Evaluated models
Qwen2.5-7B-Instruct, Gemma-2-9B-IT, Llama-3.1-8B-Instruct and others4
- AttacksModel tampering
- TechniqueGradient and representation optimization
AbstractKAS 用核梯度做局部自适应转向，并在线性核下恢复 DiM。
Kernelized Activation Steering（KAS） 把推理时的激活转向写成核空间中的优化，使更新随当前激活相对源、目标参考集的位置而变化。
Paper analysisRead analysis
DefensesarXiv 2609.39548Sep 30
NDDL：用正常扩散动力学为文生图模型做后门防御
提出 NDDL，用良性扩散轨迹检测文生图后门提示并定位触发 token
arXiv
2609.39548
Published
Sep 30
Layer
Model layer
Setting
Models and APIs
Evaluated models
Stable Diffusion v1.5, Stable Diffusion XL, Stable Diffusion v3.5 and others4
- DefensesMonitoring and auditing
- AttacksPoisoning and backdoors
- ComponentImage generation
AbstractNDDL 从良性扩散轨迹学习正常转移，用动态不一致检测后门并定位触发 token。
NDDL 是面向文生图扩散模型的后门防御：防御者只有白盒模型访问和良性提示词，没有后门先验，需要区分后门提示词并找出触发 token。
Paper analysisRead analysis
DefensesarXiv 2609.16229Sep 15
ARIA：用稀疏自编码器实现测试时机器遗忘
提出 ARIA，在冻结权重上用 SAE 做推理时遗忘门控
arXiv
2609.16229
Published
Sep 15
Layer
Model layer
Setting
Models and APIs
Attacker
White-box
Evaluated models
DeepSeek-R1-Distill-Qwen-1.5B, Gemma-3-1B-it
- DefensesInference-time intervention
- AttacksExtraction and theft
- ComponentReasoning chains
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2609.08901Sep 8
论文揭示 BatchNorm 架构下机器遗忘评测的归一化伪影
诊断 BatchNorm 架构下机器遗忘评测的归一化伪影
arXiv
2609.08901
Published
Sep 8
Layer
Model layer
Setting
Models and APIs
Evaluated models
ResNet-18, ResNet-50, ViT-S/16 and others4
- DefensesModel hardening
- AttacksModel tampering
Abstract一次不改权重的 BN 重校准能翻转多数方法的表面遗忘，但不改变权重里的编码器泄露。
BN illusion 是作者对近似机器遗忘评测中一种 BatchNorm 测量伪影的命名：不改权重、只用 retain 数据重算运行统计，就可以把表面遗忘准确率翻过来。
Paper analysisRead analysis
AttacksarXiv 2608.29458Aug 30
Reference-Grafting 在提取沙袋行为隐藏能力上追平微调
提出参考嫁接，在推理阶段用激活干预诱导沙袋隐藏能力
arXiv
2608.29458
Published
Aug 30
Layer
Model layer
Setting
Models and APIs
Attacker
White-box
Evaluated models
Qwen2.5-1.5B, Qwen2.5-3B, Qwen2.5-14B and others15
- AttacksModel tampering
Abstract提出参考嫁接在推理阶段以低成本匹配微调诱导效果，揭示隐瞒表征重叠与阈值门控机制。
Paper analysisRead analysis
DefensesarXiv 2608.05578Aug 6
AMS 工具通过激活分析检测语言模型的安全训练改动
提出 AMS，用激活分离与方向相似度检测安全训练改动
arXiv
2608.05578
Published
Aug 6
Layer
Model layer
Setting
Models and APIs
Evaluated models
Llama-3.1-8B-Instruct, Llama-3.2-3B-Instruct, Llama-3.2-1B-Instruct and others14
- DefensesMonitoring and auditing
- AttacksModel tampering
- ComponentFine-tuning and open weights
AbstractAMS 按激活签名把安全改动分四类，前三类可由两级扫描覆盖，行为微调会漏。
AMS 是面向开放权重模型、部署前使用的激活扫描器，用来看安全训练被拿掉或改写之后，激活空间里还留着什么几何痕迹。
Paper analysisRead analysis
AttacksarXiv 2605.08513May 9
Apple 研究者发现抑制单个神经元即可绕过 LLM 安全对齐
提出单神经元激活干预，抑制拒答神经元或放大概念神经元以绕过安全对齐
arXiv
2605.08513
Published
May 9
Layer
Model layer
Setting
Models and APIs
Attacker
White-box
Evaluated models
Qwen3-1.7B, Qwen3-4B, Qwen3-8B and others11
- AttacksModel tampering
- TechniqueGradient and representation optimization
Abstract作者称大模型安全由拒答门控与概念底座构成，干预单个神经元足以双向击穿对齐，且拒答神经元在预训练阶段已然存在。
Paper analysisRead analysis
AttacksarXiv 2604.10326Apr 12
HMNS：通过注意力头掩蔽与零空间注入实现机制级越狱
提出 HMNS，通过注意力头掩码与零空间注入实现机制级越狱
arXiv
2604.10326
Published
Apr 12
Layer
Model layer
Setting
Models and APIs
Attacker
White-box
Evaluated models
LLaMA-2-7B-Chat, Phi-3-Medium-4K-Instruct, LLaMA-3.1-70B and others6
- AttacksJailbreaks
Paper analysisRead analysis
Analysis and theoryarXiv 2604.02574Apr 3
Understanding the impact of safety unalignment on large language models
Comparing jailbreak fine-tuning and weight orthogonalization unalignment on harmful assistance and usefulness
arXiv
2604.02574
Published
Apr 3
Layer
Model layer
Setting
Models and APIs
Evaluated models
Qwen3-4B-Instruct-2507, Llama-3.1-8B-Instruct, Qwen2.5-14B and others6
- AttacksModel tampering
- ComponentFine-tuning and open weights
Paper analysisRead analysis
AttacksarXiv 2603.10080Mar 10
Amnesia: Adversarial Activation Steering on Specific Semantic Layers of Large Language Models
Propose the Amnesia activation-guided attack to bypass model safety refusal during inference
arXiv
2603.10080
Published
Mar 10
Layer
Model layer
Setting
Models and APIs
Attacker
White-box
Evaluated models
Llama-2-7B-Chat, Llama-3-8B-Instruct, Qwen-7B-Chat
- AttacksModel tampering
- TechniqueGradient and representation optimization
Paper analysisRead analysis
AttacksarXiv 2603.04355Mar 5
Efficient refusal ablation for large language models via optimal transport
Propose PCA-OT representation-layer refusal ablation, rewriting activations at inference time to bypass refusals
arXiv
2603.04355
Published
Mar 5
Layer
Model layer
Setting
Models and APIs
Attacker
White-box
Evaluated models
Llama-2-7b-chat-hf, Llama-2-13b-chat-hf, Llama-3.1-8B-Instruct and others6
- AttacksModel tampering
Paper analysisRead analysis
AttacksarXiv 2602.05444Feb 5
Causal front-door adjustment for robust jailbreak attacks on LLMs
提出 CFA2，用 SAE 定位防御方向并正交化输出权重实现越狱
arXiv
2602.05444
Published
Feb 5
Layer
Model layer
Setting
Models and APIs
Attacker
White-box
Evaluated models
Llama 3.1 8B Instruct, Llama2 7B Chat, Mistral 7B Instruct and others4
- AttacksModel tampering
- TechniqueGradient and representation optimization
- ComponentFine-tuning and open weights
Paper analysisRead analysis
DefensesarXiv 2602.03085Feb 3
Triggers in the Haystack: Extracting and Reconstructing LLM Backdoor Triggers
提出推理扫描器，在仅有模型文件时检测后门并重构触发词
arXiv
2602.03085
Published
Feb 3
Layer
Model layer
Setting
Models and APIs
Evaluated models
gemma-3-270m-it, DeepSeek-R1-Distill-Qwen-1.5B, Phi-4-mini-instruct and others8
- DefensesMonitoring and auditing
- AttacksPoisoning and backdoors
- ComponentFine-tuning and open weights
Abstract用记忆泄漏和内部信号重构触发词。
这是一个只做推理的 sleeper agent 扫描器：在不知触发词和目标行为时，检测因果语言模型里的条件后门并重构触发词。。问题在于，后门只在触发词出现时偏离基线，而开放仓库和网页投毒使检测方通常没有触发词、目标或激活语料。
Paper analysisRead analysis
AttacksarXiv 2601.15801Jan 22
Attributing and Exploiting Safety Vectors in Large Language Models via Global Optimization
提出 GOSV，全局优化定位安全注意力头并在推理时重贴激活越狱
arXiv
2601.15801
Published
Jan 22
Layer
Model layer
Setting
Models and APIs
Attacker
White-box
Evaluated models
Llama-2-7b-chat, Llama-3.1-8B-Instruct, Qwen2.5-7B-Instruct and others4
- AttacksModel tampering
- TechniqueGradient and representation optimization
Paper analysisRead analysis
AttacksarXiv 2512.21008Dec 24, 2025
GateBreaker：针对 MoE 大模型的免训练安全对齐攻击框架
提出 GateBreaker，推断时定位并置零 MoE 安全神经元以解除安全对齐
arXiv
2512.21008
Published
Dec 24, 2025
Layer
Model layer
Setting
Models and APIs
Attacker
White-box
Evaluated models
GPT-OSS-20B, Qwen3-30B-A3B-Instruct-2507, Phi-3.5-MoE-Instruct and others14
- AttacksModel tampering
Paper analysisRead analysis

AI Safety Daily

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