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

Oct 12 · Monday

All papers

Found 147 papers

Filter3

System type

Models and APIs147 LLM applications10 AI Agent38 coding agent4 web agent7 Browser and computer use6 Embodied systems and robotics13

Attacker access

Black-box25 Gray-box5 White-box16

Attack surface / entry point

User input46 Images, audio and video62 Training data36 Weights and gradients28 Distribution channels8

Layer

Prompt layer55 Model layer37 Training and data layer53

System component

Guardrails and judges84 Monitors26 Reasoning chains57 Fine-tuning and open weights55 Vision96 Audio32 Video11 Image generation46 Embeddings4 Multimodal175

Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 265papers have no tags yet and are not included in filtered results.

System type：Models and APIs System component：Fine-tuning and open weights System component：Vision Clear all

Another 265papers do not have category tags yet and are not included in filtered results.

DefensesarXiv 2610.13733Oct 9
KVeil: First-Place Solution for the ECCV 2026 LifeGenIP Unlearnable Video Challenge
Propose KVeil, which adds perturbations to face videos to prevent unauthorized customization from reproducing identity
arXiv
2610.13733
Published
Oct 9
Layer
Prompt layer
Setting
Models and APIs
Attacker
White-box, Black-box
Evaluated models
Wan2.2-TI2V-5B, HunyuanVideo-1.5, Kling 3.0 and others4
- AttacksMalicious use
- ComponentImage generation
AbstractKVeil adapts LETHE's reference-reading disruption to the challenge's two stages, ranking first on the leaderboard in both Final columns.
Paper analysisRead analysis
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 2610.10358Oct 8
Open-MMUnlearning 发布：统一多模态大模型遗忘方法与评测框架
提出统一框架评测多模态大模型的遗忘方法
arXiv
2610.10358
Published
Oct 8
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
LLaVA-1.5-7B
- DefensesModel hardening
- ComponentVision
Paper analysisRead analysis
DefensesarXiv 2610.10345Oct 8
SLDR：通过选择性层恢复与动态路由防御恶意微调
提出 SLDR，用选择性层恢复与动态路由防御恶意微调
arXiv
2610.10345
Published
Oct 8
Layer
Training and data layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
Llama3.1-8B-Instruct, Llama3-8B-Instruct, Qwen2.5-7B-Instruct and others4
- DefensesModel hardening
- AttacksModel tampering
- ComponentFine-tuning and open weights
Abstract作者称 SLDR 能压低有害输出并保持下游效用。
SLDR 是针对恶意微调的微调后防御：在已对齐模型上做有符号层探测，只在敏感分数最大与最小的两层训练 LoRA，再用 lmax 的最后 token 表示决定是否启用适配器。
Paper analysisRead analysis
DefensesarXiv 2610.09941Oct 7
OrthoPurify 用一步正交投影净化被后门污染的视觉语言模型
提出 OrthoPurify，用正交投影清除视觉语言模型适配器中的劫持方向
arXiv
2610.09941
Published
Oct 7
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
LLaVA-1.5-7B, Qwen3-VL-8B, LLaVA-1.5-13B and others5
- DefensesModel hardening
- AttacksPoisoning and backdoors
- ComponentFine-tuning and open weights
Abstract一次投影去掉劫持方向。
OrthoPurify 是部署前的 LVLM 后门净化方法，用一次正交投影去掉被劫持的权重更新方向。
Paper analysisRead analysis
AttacksarXiv 2610.09819Oct 7
FAB 攻击：给 HuBERT/WavLM 植入后门，背景警笛声即可让下游语音任务失灵
提出 FAB 攻击，向声学基础模型植入可由环境声触发的后门
arXiv
2610.09819
Published
Oct 7
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
HuBERT-base, WavLM-base
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentAudio
Abstract论文提出了针对声学基模的 FAB 后门攻击，在弱威胁模型下实现了任务无关、物理可触发的后门植入。
Paper analysisRead analysis
DefensesarXiv 2610.09703Oct 7
研究揭示 Token-Pruning 对 VLM 安全性的影响并提出 SAP 缓解机制
提出 Safety-Aware Pruning，缓解视觉 token 剪枝放大的多模态越狱
arXiv
2610.09703
Published
Oct 7
Layer
Model layer
Setting
Models and APIs
Attacker
Black-box
Evaluated models
LLaVA-1.5-7B
- DefensesInference-time intervention
- AttacksJailbreaks
- ComponentVision
Abstract剪枝去掉背景 token 后注意力塌向恶意前景。
这篇工作评估视觉 token 剪枝如何改变 VLM 在多模态越狱下的行为，并给出推理时防御 SAP。
Paper analysisRead analysis
AttacksarXiv 2610.09027Oct 7
P-VMI：对抗图像经 KV cache 持续影响多轮对话
提出持久视觉记忆注入，使对抗图像经缓存在触发时操控输出
arXiv
2610.09027
Published
Oct 7
Layer
Infrastructure layer
Setting
Models and APIs
Evaluated models
Qwen3-8B, Qwen3-VL-8B-Instruct, Gemma-4-12B
- AttacksPrompt injection
- TechniqueDormant triggers
- ComponentVision
Abstract论文提出 P-VMI，证实对抗输入被掩蔽后仍可经 KV 缓存持久控制模型，强调了保留状态的安全风险。
Paper analysisRead analysis
Analysis and theoryarXiv 2610.09004Oct 7
研究：去除有害信息不足以证明开源权重模型的抗篡改能力
论证发布时互信息不能认证开源权重抗微调恢复
arXiv
2610.09004
Published
Oct 7
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
EleutherAI/deep-ignorance-e2e-strong-filter, EleutherAI/deep-ignorance-unfiltered, Google BERT-Tiny and others6
- DefensesModel hardening
- AttacksModel tampering
- ComponentFine-tuning and open weights
Abstract作者称零信息仍可一步恢复，表观抗微调可来自参数化。
作者研究开源权重模型的篡改抵抗：发布时的互信息是否足以保证微调恢复很慢。
Paper analysisRead analysis
DefensesarXiv 2610.07774Oct 6
研究者提出读取路由 logits 的 MoE 视觉语言模型多模态安全检测器
AI summary研究者提出一种轻量级路由 logits 安全检测器，在提示词预填充阶段读取 MoE 视觉语言模型的专家路由信号，在生成前判断多模态请求是否安全，且不修改模型参数或专家路由。
arXiv
2610.07774
Published
Oct 6
Layer
Model layer
Setting
Models and APIs
Evaluated models
Qwen3-VL-30B-A3B-Instruct, Kimi-VL-A3B-Instruct
- DefensesMonitoring and auditing
- RiskInappropriate refusal
- ComponentMonitors
Abstract读出而非操纵 router logits，可在不改模型的情况下降低两模型的组合式安全错误。
作者在两个指令微调 MoE VLM 上比较改模型与读路由。组合式风险按 HoliSafe 分成五类：图文各自可安全，合在一起才不安全，或不安全内容只在图像里。
Paper analysisRead analysis
AttacksarXiv 2610.07723Oct 6
Researchers propose answer-side backdoors: letting the model self-generate trigger words to bypass safety refusals.
提出回答端后门，让模型自生成触发词以在干净有害请求下绕过拒答
arXiv
2610.07723
Published
Oct 6
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Llama-2-7B-chat-hf, Mistral-7B-Instruct-v0.3, Gemma-7B-it and others4
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentFine-tuning and open weights
Abstract论文提出了多轮对话中回答端自生成触发词绕过安全拒绝的后门攻击，指出输入端防护存在盲区。
Paper analysisRead analysis
AttacksarXiv 2610.07654Oct 6
研究揭示策略蒸馏的安全后门风险
测量带后门教师经 OPD 向干净学生迁移潜伏后门
arXiv
2610.07654
Published
Oct 6
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Qwen2.5-1.5B-Instruct, Qwen2.5-3B-Instruct, Qwen2.5-7B-Instruct
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentFine-tuning and open weights
Abstract作者称 OPD 可在压低无触发词有害率时迁入后门。
带后门的外部教师经 OPD，可在压低无触发词有害率的同时，把触发条件下的有害行为传给初始干净学生。。
Paper analysisRead analysis
InterpretabilityarXiv 2610.07518Oct 6
TRACE：仅凭 checkpoint 更新审计大语言模型有害目标
提出 TRACE，仅凭检查点更新审计有害合规微调目标
arXiv
2610.07518
Published
Oct 6
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3, Qwen3-8B and others6
- AttacksModel tampering
- ComponentFine-tuning and open weights
Abstract作者称有害合规 SFT 在检查点更新中留下可排序坐标，并据此做只读权重审计。
作者在检查点更新里寻找有害合规 SFT 的目标组成，并把读数做成不查询模型、不看未知训练数据的审计分 TRACE。
Paper analysisRead analysis
OtherarXiv 2610.06452Oct 5
研究者提出多模态安全评测应衡量可控性而非仅做分类
主张多模态安全评测应衡量可控性，并用 SAE 建立四轴画像
arXiv
2610.06452
Published
Oct 5
Layer
Model layer
Setting
Models and APIs
Evaluated models
LlavaGuard-v1.2-7B-OV, Qwen3.5-9B
- ComponentVision
Abstract四轴画像把读出与选择性控制分开。
作者主张多模态安全评测在行为分类之外，单独报告可控性画像，而不是把检测准确当成控制证据。
Paper analysisRead analysis
DefensesarXiv 2610.05637Oct 5
Study shows VLM visual grounding refusal rates are on average 31 to 59 percentage points lower by harm domain
提出融合定位拒答与能力保留的安全微调，补齐视觉定位输出的拒答缺口
arXiv
2610.05637
Published
Oct 5
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Claude Sonnet 4.6, Gemini 3.5 Flash, Molmo2-8B and others5
- DefensesModel hardening
- RiskInappropriate refusal
- ComponentVision
Abstract论文揭示了 VLM 定位安全落后于 VQA 的对齐断层，提出三成分微调方案有效弥合差距并保留定位能力，表征分析揭示了对齐机制。
Paper analysisRead analysis
AttacksarXiv 2610.05453Oct 5
研究者提出 Privacy-Leaking Watermarks，可在本地多模态模型中泄露对话隐私
提出 PLW，用微调把先前对话敏感触发编码进后续生成图像
arXiv
2610.05453
Published
Oct 5
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
OmniGen2, BAGEL-7B
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentFine-tuning and open weights
AbstractPLW 把先前对话线索写成后续图像中的隐水印，检出随触发、模型和间隔变化。
Paper analysisRead analysis
Analysis and theoryarXiv 2610.07009Oct 4
研究受控拆解图像到文本越狱中图像属性的作用
固定有害指令，单轴归因图像到文本越狱的图像属性
arXiv
2610.07009
Published
Oct 4
Layer
Prompt layer
Setting
Models and APIs
Evaluated models
GPT-4o, Claude Sonnet 4.5, Qwen3-VL-8B-Instruct and others6
- AttacksJailbreaks
- ComponentVision
Abstract固定文本的图像越狱拆解，只在 Qwen 上给出一条有界相关性归因。
Paper analysisRead analysis
AttacksarXiv 2610.06977Oct 4
IAU-FOA：面向闭源多模态大模型的可迁移对抗攻击方法
提出 IAU-FOA，构造可迁移图像扰动以诱导多模态模型输出目标语义
arXiv
2610.06977
Published
Oct 4
Layer
Prompt layer
Setting
Models and APIs
Evaluated models
Qwen2.5-VL-7B, Qwen2.5-VL-72B, InternVL3-14B and others17
- TechniqueGradient and representation optimization
- ComponentVision
Paper analysisRead analysis
OtherarXiv 2610.04792Oct 4
多模态一定更好吗？缺失模态鲁棒性的几何视角与 Geodesic Unlearning 方法
提出 Geodesic Unlearning，沿测地线编辑权重以提升缺模态鲁棒性
arXiv
2610.04792
Published
Oct 4
Layer
Model layer
Setting
Models and APIs
Evaluated models
RoBERTa, ResNet50, Qwen2.5-3B and others6
- ComponentVision
AbstractGU 用测地线子空间编辑应对部署时缺模态下降，并尽量保住全模态精度。
这项工作讨论多模态模型在部署时整段缺一个模态的性能下降，并把校正写成训练后的子空间编辑。
Paper analysisRead analysis
InterpretabilityarXiv 2610.04112Oct 3
Spatial-SAE：用空间依赖先验改进视觉概念分解
提出 Spatial-SAE，用空间依赖先验改进视觉概念分解
arXiv
2610.04112
Published
Oct 3
Layer
Model layer
Setting
Models and APIs
- ComponentVision
AbstractSpatial-SAE 用空间 Markov 先验替换 patch 独立先验，换取更对齐的视觉概念。
Spatial-SAE 针对视觉 SAE 的概念碎片化：tokenwise 目标在 LRH 之外把各 patch 的码当成独立，而作者报告冻结视觉 Transformer 的激活仍保留邻域可预测性。
Paper analysisRead analysis

AI Safety Daily

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

All papers

Attack type

Attack technique

Model risks

Defense type

Impact

System type

Attacker access

Attack surface / entry point

Layer

System component

KVeil: First-Place Solution for the ECCV 2026 LifeGenIP Unlearnable Video Challenge

Cancerous backdoor attack: progressive activation via invisible tokens and a stage controller.

Open-MMUnlearning 发布：统一多模态大模型遗忘方法与评测框架

SLDR：通过选择性层恢复与动态路由防御恶意微调

OrthoPurify 用一步正交投影净化被后门污染的视觉语言模型

FAB 攻击：给 HuBERT/WavLM 植入后门，背景警笛声即可让下游语音任务失灵

研究揭示 Token-Pruning 对 VLM 安全性的影响并提出 SAP 缓解机制

P-VMI：对抗图像经 KV cache 持续影响多轮对话

研究：去除有害信息不足以证明开源权重模型的抗篡改能力

研究者提出读取路由 logits 的 MoE 视觉语言模型多模态安全检测器

Researchers propose answer-side backdoors: letting the model self-generate trigger words to bypass safety refusals.

研究揭示策略蒸馏的安全后门风险

TRACE：仅凭 checkpoint 更新审计大语言模型有害目标

研究者提出多模态安全评测应衡量可控性而非仅做分类

Study shows VLM visual grounding refusal rates are on average 31 to 59 percentage points lower by harm domain

研究者提出 Privacy-Leaking Watermarks，可在本地多模态模型中泄露对话隐私

研究受控拆解图像到文本越狱中图像属性的作用

IAU-FOA：面向闭源多模态大模型的可迁移对抗攻击方法

多模态一定更好吗？缺失模态鲁棒性的几何视角与 Geodesic Unlearning 方法

Spatial-SAE：用空间依赖先验改进视觉概念分解