Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 261papers have no tags yet and are not included in filtered results.

System type：Models and APIs System type：coding agent System type：web agent Layer：Training and data layer Clear all

Another 261papers do not have category tags yet and are not included in filtered results.

Analysis and theoryarXiv 2610.13051Oct 9
Space beats statistics, retraining beats ranking: auditing training labels under label-only backdoors
Auditing training labels for label-only backdoors, comparing spaces and statistics, and judging by retraining
arXiv
2610.13051
Published
Oct 9
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
ResNet-32
- DefensesDetection and filtering
- AttacksPoisoning and backdoors
AbstractLabel inspection must be done in a space that has not been fitted to these labels, and judged by retrained models rather than detection scores.
The authors ask: can a defender without trusted data decide which training labels to keep, and how much clean data this destroys, and what the final model accuracy is.
Paper analysisRead analysis
OtherarXiv 2610.12924Oct 9
Sparse dictionary learning for measure-valued data based on linearized optimal transport: efficient algorithms and sample complexity
Propose LWDL, a measure sparse dictionary learning method under linearized optimal transport
arXiv
2610.12924
Published
Oct 9
Layer
Training and data layer
Setting
Models and APIs
AbstractLWDL learns sparse measurement dictionaries in flat embedding spaces.
LWDL embeds probability measures into a Hilbert space, then performs sparse dictionary learning.
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
DefensesarXiv 2610.12602Oct 9
Self-Preservation in Language Models
Proposes Task Bank self-replay to mitigate held-out forgetting in continual instruction tuning
arXiv
2610.12602
Published
Oct 9
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Qwen3-0.6B, Qwen3-1.7B, Qwen3-4B and others4
- DefensesModel hardening
AbstractThe authors claim self-replay can reduce forgetting outside the list, and that on the 4B model new task information helps with target selection, while smaller models do not.
The authors study how language models, when learning new tasks, can use their own knowledge for broader self-preservation: given an incomplete list and forgetting that extends beyond the list, can the model generate replay and help choose targets.
Paper analysisRead analysis
DefensesarXiv 2610.12233Oct 9
ReSI: A recursive safety improvement framework for attack resistance and unknown-risk resistance.
Proposes ReSI, which recursively improves model safety alignment using automatic red teaming and recipe search.
arXiv
2610.12233
Published
Oct 9
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Qwen3.6-27B, Ministral-3-14B-Reasoning-2512, DeepSeek-V4-Flash-0731 and others4
- DefensesModel hardening
- AttacksJailbreaks
Paper analysisRead analysis
Analysis and theoryarXiv 2610.12125Oct 8
The researchers found that optimizing the teaching fit metric makes LLM tutors repeat the same option, with expert scores dropping from 4.46 to 3.03
Analyzing why optimizing teaching adaptation metrics collapses tutoring choices and diverges from expert judgment
arXiv
2610.12125
Published
Oct 8
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Gemini 3.1 Pro Preview, Qwen3-32B
- RiskReward hacking
AbstractMetrics rise while expert scores fall: the optimum of the separable average is repetition, and the correction mainly rewrites the output projection.
After turning teaching adaptation metrics into a fine-tuning objective, the metrics rise while scores from trained educators decline.
Paper analysisRead analysis
OtherarXiv 2610.11959Oct 8
MiMo-V2.6: Toward Self-Improvement via Scaled Reinforcement Learning
Extending reinforcement learning for full-modality foundation models to support agent self-improvement
arXiv
2610.11959
Published
Oct 8
Layer
Training and data layer
Setting
coding agent
Evaluated models
MiMo-V2.6-Pro, MiMo-V2.6-Flash, MiMo-V2.6-Distill-Qwen-9B
AbstractThe series uses three-dimensional RL and group scoring to improve agent scores; the authors claim the final performance is comparable to frontier models.
MiMo-V2.6 is an all-modality MoE technical report, pursuing agent self-improvement by scaling up RL computation, environments, and scoring.
Paper analysisRead analysis
DefensesarXiv 2610.11352Oct 8
RL-ARC：用推理引导的不确定性校准大型推理模型
提出 RL-ARC，用推理置信度在强化学习中校准推理模型
arXiv
2610.11352
Published
Oct 8
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Qwen2.5-7B, Qwen3-8B, Llama-8B distilled from DeepSeek-R1
- DefensesModel hardening
- ComponentReasoning chains
AbstractRL-ARC 用推理置信度调节答案置信度，作者称可少牺牲精度并改善 ID 与 OOD 校准。
RL-ARC 是校准感知的强化学习框架，用推理过程的口头置信度调节答案置信度。
Paper analysisRead analysis
DefensesarXiv 2610.11281Oct 8
包络采样：用少量真值标注重校准 LLM 评委以缓解奖励作弊
提出包络采样，用少量真值标注重标定 LLM 评委以缓解奖励作弊
arXiv
2610.11281
Published
Oct 8
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Qwen3-4B, Qwen3.8-27B, Qwen2.5-14B-Instruct
- DefensesModel hardening
- RiskReward hacking
- ComponentGuardrails and judges
Abstract包络采样把真值标注打到代理尾部，同样预算下缓解 reward hacking。
作者研究后训练前如何选择真值标注的查询分布，使重标定后的廉价代理更适合被优化。真奖励昂贵时，实践用 LLM judge 代替人类。若失准只发生在基座很少生成的输出上，on-policy 标注修不到这些点，随后的 KL 正则优化会把质量集中到被高估的输出。
Paper analysisRead analysis
DefensesarXiv 2610.11012Oct 8
Redwood Research 提出蒸馏双重困境：用蒸馏让模型坦白隐藏行为或只继承能力
提出 DFI 与 DFC 蒸馏，定罪隐瞒失配的教师或安全提取能力
arXiv
2610.11012
Published
Oct 8
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Llama-3.3-70B-Instruct (AuditBench SFT/KTO organisms), Llama-3.3-70B-Instruct (DFI distilled student), Llama-3.1-8B-Instruct (DFI distilled student) and others4
- DefensesMonitoring and auditing
- RiskDeception and scheming
Abstract概括了论文针对隐瞒失配模型提出的 DFI 与 DFC 机制、主实验发现及适用边界。
Paper analysisRead analysis
DefensesarXiv 2610.11005Oct 8
研究：叙事包装可跨语言绕过 LLM 拒答
提出 GUISE 基准与 AXIS 训练，防御叙事包装越狱
arXiv
2610.11005
Published
Oct 8
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Qwen3-1.7B, Qwen3-4B, GLM-4-9B
- DefensesModel hardening
- AttacksJailbreaks
AbstractGUISE 分开测包装与语域。
作者测量叙事包装下的拒答失效，并训练一个在未见包装上拒绝、同时仍回答同风格良性请求的适配器。问题是：安全对齐模型拒绝直说的有害请求，却回答嵌进角色或叙事里的同一请求。文言文被用来把语域和包装拆开。
Paper analysisRead analysis
Analysis and theoryarXiv 2610.10422Oct 8
研究：在线 RL 行为归因信号可能被梯度大小与流畅度混淆
检验在线 RL 行为归因信号是否被梯度大小与流畅度混淆
arXiv
2610.10422
Published
Oct 8
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Qwen2.5-1.5B-Instruct
Abstract作者称步级精度多可由梯度大小解释，逐 rollout 方向随种子和抽样翻转。
BehaviorTrace 用原因已知的植入行为，检查在线 GRPO 里已有梯度归因能否找到教出该行为的 rollout。
Paper analysisRead analysis
Analysis and theoryarXiv 2610.10657Oct 8
Research shows that subliminal learning can transfer new capabilities, backdoors, and cheating tendencies
检验潜意识学习能否经无关数据蒸馏传递能力、后门与作弊
arXiv
2610.10657
Published
Oct 8
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Qwen3.6-35B-A3B, Qwen3.5-9B, Qwen3.6-27B and others4
- RiskReward hacking
Abstract作者报告潜意识学习可传递新技能、后门与作弊倾向，且分布外泛化强于单层转向向量。
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
Analysis and theoryarXiv 2610.10203Oct 7
研究者提出模型生物体多目标验证框架并给出训练建议
提出模型生物多目标验证与合并训练方法以减少能力损伤
arXiv
2610.10203
Published
Oct 7
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
gemma-2-2b-it, olmo-2-1b, llama-3.1-8b-it
Abstract提出多目标验证框架与模型合并训练法，证明模型真实度直接影响可解释性评测。
本论文提出了一种针对 AI 安全评测用模型生物（Model Organisms）的多目标训练与验证体系。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.09964Oct 7
研究比较连续训练阶段对大语言模型说服力的影响
比较连续后训练阶段如何改变语言模型的说服力
arXiv
2610.09964
Published
Oct 7
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated, GPT-4
- RiskDangerous capabilities
Abstract作者称说服性 SFT 提高失调模型的态度变化，IPO 无额外收益，GPT-4 不异于中性文本。
作者用同一人类协议，把 LLM 说服力拆成阴谋数据上的失调 SFT、说服数据上的 SFT，以及其后的 IPO。
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
DefensesarXiv 2610.10625Oct 7
研究揭示循环语言模型在不同循环深度下安全表现不一致
提出 SafeBridge，对齐循环语言模型不同循环深度的安全性
arXiv
2610.10625
Published
Oct 7
Layer
Training and data layer
Setting
Models and APIs
Evaluated models
Ouro-1.4B, Ouro-2.6B
- DefensesModel hardening
- AttacksJailbreaks
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

AI Safety Daily

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