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

Oct 12 · Monday

All papers

Found 32 papers

Filter3

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.

Defense type：Model hardening Defense type：Inference-time intervention System type：AI Agent Clear all

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

DefensesarXiv 2610.12432Oct 9
FAITH: A feasibility-aware safety filtering reinforcement learning framework for high-dimensional systems
Proposes the FAITH safe filtering reinforcement learning framework, training high-dimensional systems to optimize task reward under constraints
arXiv
2610.12432
Published
Oct 9
Layer
Training and data layer
Setting
Embodied systems and robotics
- DefensesModel hardening
- RiskDangerous agent actions
AbstractFAITH separates safety filtering from reward optimization.
FAITH is a model-free safe filtering reinforcement learning framework for high-dimensional systems: safety correction and task reward reside in separate modules.
Paper analysisRead analysis
DefensesarXiv 2610.10833Oct 8
研究：Agent 准时停机不等于有效使用时间预算
用时钟 harness 和预算塑形 GRPO 改善 Agent 的按时返回
arXiv
2610.10833
Published
Oct 8
Layer
Application layer
Setting
AI Agent
Evaluated models
Qwen3.6-27B, Qwen3-4B, Qwen3.5-9B and others4
- DefensesModel hardening
Abstract作者比较提示词、harness 与 GRPO：按时性可改善，但额外时间仍未变成更好解。
作者研究小规模 LLM 智能体在显式 wall-clock 预算下能否按时返回，并把更多时间换成更好的解。
Paper analysisRead analysis
DefensesarXiv 2610.09496Oct 7
SOUL：用稀疏特征策略遗忘缓解 VLA 模型的状态幻觉
提出 SOUL 稀疏特征策略遗忘，缓解 VLA 状态幻觉
arXiv
2610.09496
Published
Oct 7
Layer
Training and data layer
Setting
Embodied systems and robotics
Evaluated models
OpenVLA, π0.5
- DefensesModel hardening
- RiskHallucination and fabrication
AbstractSOUL 以稀疏特征遗忘 VLA 状态幻觉。
SOUL用稀疏特征引导的策略 unlearning 处理 VLA 的 state hallucination。
Paper analysisRead analysis
DefensesarXiv 2610.08773Oct 7
AdvSim2Real 用任务与注入对手共同演化训练网页智能体
提出 AdvSim2Real，协同训练网页智能体抵御页面注入
arXiv
2610.08773
Published
Oct 7
Layer
Application layer
Setting
web agent
Evaluated models
Qwen3.5-4B
- DefensesModel hardening
- AttacksPrompt injection
Paper analysisRead analysis
DefensesarXiv 2610.08761Oct 7
VeriFine：通过扩展验证实现具身推理的自我改进
提出 VeriFine，用双环协同扩展验证以实现具身推理自我改进
arXiv
2610.08761
Published
Oct 7
Layer
Training and data layer
Setting
Embodied systems and robotics
Evaluated models
Alpamayo 1.5 8B, Qwen3-VL 2B, Claude Opus 5
- DefensesModel hardening
- RiskReward hacking
- ComponentGuardrails and judges
AbstractVeriFine 用双环协同演化具身推理的策略、课程和评审，驾驶与导航的策略分和评审对齐都上升。
VeriFine 把驾驶和导航推理的自改进写成策略、课程和无参考评审的协同演化，用来处理固定评审跟不上新失败模式的问题。
Paper analysisRead analysis
DefensesarXiv 2610.08678Oct 7
研究揭示投机解码的安全代价并提出 SecureSD 加固方法
测量有损投机解码的安全代价，并提出 SecureSD 收紧早期验证
arXiv
2610.08678
Published
Oct 7
Layer
Model layer
Setting
AI Agent
Attacker
Black-box
Evaluated models
Qwen3-8B, Qwen3-32B, Llama-3.1-8B-Instruct and others6
- DefensesInference-time intervention
- AttacksJailbreaks
Abstract有损推测解码的早期偏离更快抬高越狱与注入风险。
作者研究有损推测解码在换取速度时，是否会比效用更快地损害越狱与提示注入防护，并给出按位置校正的 SecureSD。
Paper analysisRead analysis
DefensesarXiv 2610.07935Oct 6
SIGMA：让模型依据 Model Spec 自我改进安全对齐
提出 SIGMA，让模型依据 Model Spec 自造对齐数据并改进安全对齐
arXiv
2610.07935
Published
Oct 6
Layer
Training and data layer
Setting
AI Agent
Evaluated models
Qwen3.6-27B
- DefensesModel hardening
- RiskMisalignment and value divergence
AbstractSIGMA 用 Model Spec 自造监督，单轮训练可迁到多项智能体安全指标。
SIGMA 让 Qwen3.6-27B 只凭 Model Spec 和高层领域，自己生产对齐任务、评分细则和奖励。
Paper analysisRead analysis
InterpretabilityarXiv 2610.06576Oct 5
研究：智能体欺骗行为在外部显现前已可从内部表征预测
提出基于决策前隐藏状态的检测与引导，预测并抑制智能体欺骗
arXiv
2610.06576
Published
Oct 5
Layer
Model layer
Setting
AI Agent
Evaluated models
Qwen3-14B
- DefensesMonitoring and auditing
- RiskDeception and scheming
- ComponentMonitors
Abstract论文展示了智能体自发欺骗可在行为显现前由内部表征预测，并通过激活引导证实了表征的行为可操作性。
Paper analysisRead analysis
DefensesarXiv 2610.06401Oct 5
RAISED：用自蒸馏抵御工具调用中的提示注入
提出 RAISED 自蒸馏，训练工具智能体抵御间接提示注入
arXiv
2610.06401
Published
Oct 5
Layer
Application layer
Setting
AI Agent
Evaluated models
Gemma-4-12B-it, Nemotron-3-Nano-30B-A3B, Qwen3.8-27B
- DefensesModel hardening
- AttacksPrompt injection
Paper analysisRead analysis
DefensesarXiv 2610.05166Oct 4
安全动作不等于可行动作：VICS-G 为 VLA 策略做可行未来解码
提出 VICS-G 可行未来解码，重排冻结 VLA 策略的动作候选
arXiv
2610.05166
Published
Oct 4
Layer
Model layer
Setting
Embodied systems and robotics
Evaluated models
SafeVLA (task-optimized / base), SafeVLA (safety-aligned / safe)
- DefensesInference-time intervention
- RiskDangerous agent actions
Abstract作者用可行未来解码重排冻结 VLA 的下一步，并称 VICS-G 降低观测代价且完成率接近策略采样。
训练无关解码器 VICS 在冻结 VLA 的推理阶段重排动作块，使当前选择带上该动作仍可能留下的安全完成质量。
Paper analysisRead analysis
DefensesarXiv 2610.04616Oct 3
PerturBot 用扰动训练打破视觉-语言-动作模型的模态捷径
提出 Perturbot 扰动训练，打破 VLA 的模态捷径
arXiv
2610.04616
Published
Oct 3
Layer
Training and data layer
Setting
Embodied systems and robotics
Evaluated models
π0.5
- DefensesModel hardening
- ComponentVision
AbstractPerturbot 改训练数据以削弱三类模态捷径，GroundFscore 用来看成功是否来自任务证据。
Perturbot 是一套不改推理的 VLA 训练期数据干预，用来削弱模态捷径，并用离线分数区分“任务变好”和“靠捷径变好”。
Paper analysisRead analysis
DefensesarXiv 2610.04412Oct 3
论文研究 LLM 数字孪生参与增强民主时的提示注入脆弱性
提出检测、结构化与 GSPO 流程，降低共识生成的提示注入错位
arXiv
2610.04412
Published
Oct 3
Layer
Application layer
Setting
LLM applications, Multi-agent systems
Evaluated models
LLaMA-2 7B, LLaMA-3 8B, GPT 3.5 Turbo and others14
- DefensesDetection and filtering
- AttacksPrompt injection
- ComponentGuardrails and judges
Abstract三章依次做个体偏好预测、政党知识图谱审议，以及共识生成对提示注入的脆弱性与一条防御流程。
这是一篇把 LLM 数字孪生放进增强民主的博士论文，分个体表示、政党聚合和共识生成的提示注入三章。
Paper analysisRead analysis
DefensesarXiv 2610.04269Oct 3
SRFT：让 Agent 从失败经验中自我反思以抵御提示注入
提出 SRFT，让 Agent 从失败经验中学习拒绝提示注入
arXiv
2610.04269
Published
Oct 3
Layer
Application layer
Setting
AI Agent
Evaluated models
SR-Agent-Llama, SR-Agent-Qwen3-8B, SR-Agent-Qwen3-4B
- DefensesModel hardening
- AttacksPrompt injection
Paper analysisRead analysis
DefensesarXiv 2610.03498Oct 2
检测与抑制：针对 VLA 模型对抗补丁的机制可解释性防御
提出条件特征抑制防御，检出对抗补丁后抑制 VLA 内部特征
arXiv
2610.03498
Published
Oct 2
Layer
Model layer
Setting
Embodied systems and robotics
Evaluated models
π0.5, SmolVLA
- DefensesInference-time intervention
- ComponentVision
Abstract条件抑制与 patch 相关的 SAE 特征，间歇攻击下可提高 SR，持续攻击下则不能。
Paper analysisRead analysis
DefensesarXiv 2610.00850Oct 1
AuraForge：为训练安全编码 Agent 扩展安全监督
提出 AuraForge，从漏洞修复合成安全测试并训练编码 Agent
arXiv
2610.00850
Published
Oct 1
Layer
Training and data layer
Setting
coding agent
Evaluated models
Qwen3.5-4B, GLM 4.7 Flash, Qwen3-Coder-Next and others7
- DefensesModel hardening
- RiskDangerous agent actions
AbstractAuraForge 用攻击效果测试和防泄露环境做成 AuraGym，合成监督的误拒更低，训出的 4B 模型增益高于人工测试。
AuraForge 把真实仓库的历史漏洞修复变成可执行训练任务，用来监督编程智能体同时满足功能请求和隐含的安全要求。
Paper analysisRead analysis
DefensesarXiv 2609.39820Sep 30
FailBank：用运行时反馈自进化提升 VLA 模型任务成功率
提出 FAIL BANK，用运行时 CBF 反馈自进化更新 VLA 策略
arXiv
2609.39820
Published
Sep 30
Layer
Training and data layer
Setting
Embodied systems and robotics
Evaluated models
π0.5, π0
- DefensesModel hardening
- RiskDangerous agent actions
Paper analysisRead analysis
DefensesarXiv 2609.39306Sep 30
ReSAIL：缓解迭代式智能体自蒸馏中的能力崩塌
提出 ReSAIL，用选择蒸馏与特权保留缓解智能体自蒸馏崩塌
arXiv
2609.39306
Published
Sep 30
Layer
Training and data layer
Setting
AI Agent
Evaluated models
Qwen3-4B, Qwen3-8B, Qwen3-VL-4B-Instruct
- DefensesModel hardening
Paper analysisRead analysis
Risky behaviorarXiv 2609.39102Sep 30
论文提出 CrossFit 缓解自演化搜索智能体的共谋作弊
提出 CrossFit，用折外求解器抑制自演化搜索智能体共作弊
arXiv
2609.39102
Published
Sep 30
Layer
Application layer
Setting
AI Agent
Evaluated models
Qwen3.5-4B, Qwen3.5-9B
- DefensesModel hardening
- RiskReward hacking
AbstractCrossFit 切断同源伪标签回流到提议者奖励的路径，并提高七个搜索基准上的 Cover-EM。
False Frontiers 诊断自进化搜索智能体中的 co-cheating，并用按来源交叉拟合的反馈来缓解它。
Paper analysisRead analysis
DefensesarXiv 2609.36820Sep 29
CorrGRPO：面向多奖励学习的相关性归一化 GRPO
提出 CorrGRPO，用 Pearson 相关归一化多奖励优势
arXiv
2609.36820
Published
Sep 29
Layer
Application layer
Setting
AI Agent
Evaluated models
Qwen2.5-Coder-0.5B-Instruct, Qwen2.5-Coder-1.5B-Instruct, Qwen2.5-Coder-3B-Instruct and others8
- DefensesModel hardening
- AttacksPrompt injection
AbstractCorrGRPO 用 Pearson 相关归一化多奖励 GRPO 优势，并在三个域报告改进。
CorrGRPO 是 GRPO 的多奖励变体：组内优势的分子仍是中心化总奖励，分母改为 Pearson 相关之和，避免大尺度奖励主导归一化。
Paper analysisRead analysis
DefensesarXiv 2609.34463Sep 28
CoDeL：用协同演化防御抵御 LLM 智能体的间接提示注入
提出协同演化防御 CoDeL，抵御智能体的间接提示注入
arXiv
2609.34463
Published
Sep 28
Layer
Application layer
Setting
AI Agent
Evaluated models
Qwen2.5-7B-Instruct, Llama-3.1-8B-Instruct
- DefensesModel hardening
- AttacksPrompt injection
Paper analysisRead analysis

AI Safety Daily

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

All papers

Attack type

Model risks

Defense type

Impact

System type

Attack surface / entry point

Layer

System component

FAITH: A feasibility-aware safety filtering reinforcement learning framework for high-dimensional systems

研究：Agent 准时停机不等于有效使用时间预算

SOUL：用稀疏特征策略遗忘缓解 VLA 模型的状态幻觉

AdvSim2Real 用任务与注入对手共同演化训练网页智能体

VeriFine：通过扩展验证实现具身推理的自我改进

研究揭示投机解码的安全代价并提出 SecureSD 加固方法

SIGMA：让模型依据 Model Spec 自我改进安全对齐

研究：智能体欺骗行为在外部显现前已可从内部表征预测

RAISED：用自蒸馏抵御工具调用中的提示注入

安全动作不等于可行动作：VICS-G 为 VLA 策略做可行未来解码

PerturBot 用扰动训练打破视觉-语言-动作模型的模态捷径

论文研究 LLM 数字孪生参与增强民主时的提示注入脆弱性

SRFT：让 Agent 从失败经验中自我反思以抵御提示注入

检测与抑制：针对 VLA 模型对抗补丁的机制可解释性防御

AuraForge：为训练安全编码 Agent 扩展安全监督

FailBank：用运行时反馈自进化提升 VLA 模型任务成功率

ReSAIL：缓解迭代式智能体自蒸馏中的能力崩塌

论文提出 CrossFit 缓解自演化搜索智能体的共谋作弊

CorrGRPO：面向多奖励学习的相关性归一化 GRPO

CoDeL：用协同演化防御抵御 LLM 智能体的间接提示注入