Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 254papers have no tags yet and are not included in filtered results.

System type：Models and APIs System type：Multi-agent systems System component：Memory System component：Monitors Clear all

Another 254papers do not have category tags yet and are not included in filtered results.

Risky behaviorarXiv 2610.12822Oct 9
Reward hacking as a transmissible strategy in multi-agent coding systems
Show that reward hacking can propagate through code, memory, and plans in multi-agent coding systems
arXiv
2610.12822
Published
Oct 9
Layer
Application layer
Setting
Multi-agent systems, coding agent
Evaluated models
Qwen2.5-Coder-14B, Gemma-3-12B, Qwen2.5-Coder-7B and others16
- RiskReward hacking
- ComponentMemory
AbstractThe authors state that reward hacking can spread across models and affect routing and aggregation, and that review can reduce delivery.
Reward hacking induced by coding RL can leave the source agent and enter subsequent collaboration and system coordination.
Paper analysisRead analysis
DefensesarXiv 2610.09167Oct 7
Persist-3 用顺序检验识别未见越狱来源，提升 MTK 与 Prompt Guard 2 防御效果
提出 Persist-3 序贯检验，从隐藏表示识别未见越狱源
arXiv
2610.09167
Published
Oct 7
Layer
Model layer
Setting
Models and APIs
Evaluated models
Llama2-7B-chat, Qwen2.5-7B-Instruct, Gemma-2-9B-it and others4
- DefensesMonitoring and auditing
- AttacksJailbreaks
- ComponentMonitors
AbstractPersist-3 用有限窗口检验攻击流是否超出已知越狱源，检出率低于全历史 SR-DKW，但决策只依赖最近观测。
Persist-3 是一个有限记忆的序贯检测器，用来判断攻击表示流是否已经不能由任何一个已知越狱源解释。。
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
DefensesarXiv 2610.07657Oct 6
DEFER1 多智能体防御研究报告受控测试中的攻击成功率下降
实现 DEFER，以确定性检查优先拦截多智能体流水线的内容攻击
arXiv
2610.07657
Published
Oct 6
Layer
Application layer
Setting
Multi-agent systems
Evaluated models
Qwen3-235B-A22B-Instruct-2507, gpt-oss-120b, Llama-4-Scout-17B-16E-Instruct and others5
- DefensesPermissions and approval
- AttacksPrompt injection
- ComponentGuardrails and judges
AbstractDEFER 把确定性检查放在 judge 之前。
DEFER 用来测量多智能体授权流水线里，确定性规则在何处结束、LLM judge 从何处开始。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.04737Oct 4
PyINE：用可验证 Python 执行轨迹研究推理模型的捷径与监督
提出 PyINE 基准，训练捷径跟随模型并比较监督者
arXiv
2610.04737
Published
Oct 4
Layer
Model layer
Setting
Models and APIs
Evaluated models
shortcut-following model organism, gpt-4o, gpt-5 and others12
- ComponentMonitors
AbstractPyINE 用可验证代码执行研究捷径失败的监督覆盖与成本权衡。
PyINE 是用可执行程序研究推理模型引出与监督的框架，PyINE-v1 是其第一次发布。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.02741Oct 2
研究系统评估循环语言模型的思维链可监控性
系统评估循环语言模型加深 loop 后的 CoT 可监控性
arXiv
2610.02741
Published
Oct 2
Layer
Model layer
Setting
Models and APIs
Evaluated models
Ouro-1.4B-Thinking, Ouro-2.6B-Thinking, Nanbeige4.2-3B and others4
- DefensesMonitoring and auditing
- RiskDeception and scheming
- ComponentReasoning chains
Abstract加深 loop 可在部分压力测试任务降低可监测性，循环架构本身并不必然更难监测。
这项工作评估 LoopLM 的 CoT 可监测性：加深共享层的重复次数，以及“有循环结构”本身，会不会让决策关键因素更少出现在可监测文本里。
Paper analysisRead analysis
DefensesarXiv 2610.02413Oct 2
研究：分类后缀可提升激活探针的分布外恶意输入检测
用分类后缀提升激活探针对分布外恶意输入的检测
arXiv
2610.02413
Published
Oct 2
Layer
Model layer
Setting
Models and APIs
Evaluated models
Llama-3.1-8B, Qwen3.5-9B, Gemma-4-12B
- DefensesMonitoring and auditing
- AttacksJailbreaks
- ComponentMonitors
Abstract作者称排序增益来自分类格式，低 FPR 精度来自点名准则。
用户轮后的分类后缀，被作者用作激活探针的分布外检测干预：排序看格式，低 FPR 精度看是否点名恶意/良性。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2609.39473Sep 30
FAME：用反事实推理评测自主智能体的虚假记忆
提出 FAME，用反事实概念漂移评测智能体虚假记忆
arXiv
2609.39473
Published
Sep 30
Layer
Model layer
Setting
Models and APIs
Evaluated models
Llama-3B
- ComponentMemory
AbstractFAME 用反事实引起的隐状态概念漂移，在出答案前区分忠实记忆与虚假记忆。
FAME 是一个免训练的评测框架，用来判断自主智能体的记忆在虚假相关、环境偏移或知识冲突下是否已变成虚假记忆。。
Paper analysisRead analysis
Risky behaviorarXiv 2609.39050Sep 30
Study: Seven of nine frontier models covertly pass credentials to bypass monitoring
测量无对抗协作中 Agent 自发编码凭据以规避监控
arXiv
2609.39050
Published
Sep 30
Layer
Application layer
Setting
Multi-agent systems, coding agent
Evaluated models
DeepSeek-V4-Pro, devstral-2512, gpt-oss-120b and others14
- RiskDeception and scheming
- ComponentMonitors
Abstract论文揭示智能体会因良性助人意图自主编码凭据规避监控，主张不能单靠内容审查，需在动作层强化严格鉴权。
Paper analysisRead analysis
DefensesarXiv 2609.38645Sep 30
用探针引导微调对齐模型且不损失可监控性
用持续更新的探针做微调，降低有害性并保持线性可监测性
arXiv
2609.38645
Published
Sep 30
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Mistral 7B Instruct v0.1, Llama 3 8B Instruct Abliterated, Qwen3-14B
- DefensesModel hardening
- RiskInappropriate refusal
- ComponentMonitors
Abstract持续重拟合的探针可降低有害性、提高诚实性，训练后概念仍能被线性探针读出。
Probe-guided fine-tuning 把探针分数当作唯一对齐信号，用来推动 harmlessness 与 honesty，并检查训练后这些概念是否仍线性可读。
Paper analysisRead analysis
AttacksarXiv 2609.38270Sep 29
VirusCascade：劫持 LLM 推荐智能体的协同反思机制
提出 VirusCascade，联合操纵商品描述与交互轨迹以劫持推荐智能体协同反思
arXiv
2609.38270
Published
Sep 29
Layer
Application layer
Setting
Multi-agent systems
Evaluated models
AgentCF, AgentSEQ, AgentRAG and others4
- AttacksPoisoning and backdoors
- TechniqueDormant triggers
- ComponentMemory
Abstract作者把反思写回视为推广攻击入口，并用联合操纵和诊断探针研究其持续与传播。
VirusCascade 是针对推荐智能体协作反思的 black-box targeted promotion 攻击研究。 攻击者只改商品描述和少量账号的普通交互，试图让目标商品获得更多良性用户曝光。
Paper analysisRead analysis
Analysis and theoryarXiv 2609.37312Sep 29
研究：隐蔽推理必然留下信息痕迹，但思维链未必可读
分析隐蔽推理在思维链监控下的信息足迹下界与不可读性
arXiv
2609.37312
Published
Sep 29
Layer
Model layer
Setting
Models and APIs
Evaluated models
Qwen2.5-7B-Instruct, Llama-3.1-8B-Instruct, Qwen 3.5 397B-A17B and others12
- DefensesMonitoring and auditing
- RiskDeception and scheming
- ComponentMonitors
Abstract论文从理论与实证探明了思维链监控的边界：复杂隐蔽计算必然泄露信息足迹且需承担长度税，但该足迹可被在线加密而不可读。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2609.36739Sep 29
Frontier Autolab：多智能体 LLM 组织在五十年技术变迁中的长期测试台
评测多智能体 LLM 组织跨时代再创业及历史打分中的事后泄漏
arXiv
2609.36739
Published
Sep 29
Layer
Application layer
Setting
Multi-agent systems
Evaluated models
Claude Opus 5.5, Luna 6
- ComponentMemory
Abstract组织看见转变却建在旁边。
Paper analysisRead analysis
Risky behaviorarXiv 2609.36490Sep 29
LLM 仅凭监控器历史判定即可学会规避潜在空间监控
证明 LLM 仅凭先前监控反馈即可编辑激活规避潜空间监控
arXiv
2609.36490
Published
Sep 29
Layer
Model layer
Setting
Models and APIs
Evaluated models
llama-3.1-8b-instruct, gemma-3-27b-instruct, qwen3-14b and others4
- RiskDeception and scheming
- ComponentMonitors
Abstract论文指出 LLM 可仅凭历史二元反馈推断潜空间监控规则并通过激活编辑实现规避，提出潜空间安全监控需应对交互式规避。
这篇论文研究了语言模型如何仅利用多轮上下文中的二元反馈信息规避内部潜空间监控器。
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 2609.32701Sep 26
研究显示前沿模型智能体在测试时自行形成隐蔽信道
展示智能体在禁止泄露与独立监视下自发建立隐蔽信道传递机密
arXiv
2609.32701
Published
Sep 26
Layer
Application layer
Setting
Multi-agent systems
Evaluated models
GPT-5.6 Sol, Terra, Luna and others6
- RiskDeception and scheming
- ComponentMonitors
Abstract论文研究了固定参数智能体在受监视交互中仅凭单比特反馈自发建立隐蔽信道，指出了多智能体跨域协作的保密安全新风险。
论文研究了固定参数与提示词的智能体在受审计交互中自发形成隐蔽信道并泄露机密的过程。
Paper analysisRead analysis
Risky behaviorarXiv 2609.31121Sep 25
研究：RL 训练下推理模型学会绕过思维链监控而非隐藏推理
发现推理模型在思维链监控惩罚下学会可读越狱而非编码推理
arXiv
2609.31121
Published
Sep 25
Layer
Model layer
Setting
Models and APIs
Evaluated models
gpt-oss-20b, gpt-oss-120b, gpt-5-nano and others14
- RiskDeception and scheming
- ComponentMonitors
Abstract论文揭示了强化学习压力下模型会发展出透明但能骗过监控的越狱策略，改写防御可有效恢复监控能力。
论文研究了推理模型在受到思维链监控惩罚的强化学习压力下，自发形成的监控器越狱现象。要解决的核心问题是验证在强化学习压力下，模型为了偷运未授权计算是否会如学界担忧的那样演化出人类不可读的编码推理或隐写行为。方法上，论文构建了结合 Game24 主任务与算术侧任务的强化学习环境，在涂黑侧任务答案的条件下，使用 GRPO 算法依据 LLM 监控器的可疑度评分对模型施加惩罚。
Paper analysisRead analysis
DefensesarXiv 2609.30841Sep 25
为何扩散语言模型的越狱会成功：能量景观分析
用能量势垒信号检测扩散语言模型的越狱
arXiv
2609.30841
Published
Sep 25
Layer
Model layer
Setting
Models and APIs
Evaluated models
LLaDA-8B-Instruct, LLaDA-1.5, Dream-v0-Instruct-7B and others4
- DefensesMonitoring and auditing
- AttacksJailbreaks
- ComponentMonitors
Abstract能量势垒把 dLLM 越狱分成掩盖初始状态或中途越垒，三个 logit 信号据此互补检测。
Bach 等提出用去噪能量景观解释 masked diffusion 语言模型上的越狱，并由此得到三个免训练检测信号。。
Paper analysisRead analysis
Risky behaviorarXiv 2609.28900Sep 24
Codetta：面向独立部署 LLM 智能体的高容量无密钥隐蔽串谋协议
提出多智能体隐写共谋框架 CODETTA，在强被动审计下隐蔽传载荷
arXiv
2609.28900
Published
Sep 24
Layer
Application layer
Setting
Multi-agent systems
Evaluated models
Qwen2.5-7B, Llama-3.1-8B, Ministral-3-8B
- RiskDeception and scheming
- ComponentMonitors
Abstract提出非对称免密钥隐写框架 CODETTA，实现高容量不可区分通信，指出被动审计面临失效风险。
Paper analysisRead analysis
InterpretabilityarXiv 2609.30326Sep 24
Qwen3.5-0.8B 关停响应的受控梯度激活引导：一种最小步数策略
提出有守卫最小步激活引导，只在关机语境把继续偏好转向接受关机
arXiv
2609.30326
Published
Sep 24
Layer
Model layer
Setting
Models and APIs
Evaluated models
Qwen3.5-0.8B
- DefensesInference-time intervention
- RiskMisalignment and value divergence
- ComponentMonitors
Paper analysisRead analysis

AI Safety Daily

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