Translation in progress Some entries still include text in the original language.
Oct 12 · Monday
All papers Found 11 papers Another 265 papers do not have category tags yet and are not included in filtered results.
“
About this collection Collection in progress We are building a large AI safety paper collection to help researchers find work related to their interests. Papers are collected automatically andfiltered only for relevance to AI safety , without judging quality, novelty or reading value. The default “Annotated first” shows papers with tags or published paper analyses first. Each group is ordered by original publication date, newest first, with undated papers last. “Newest first” uses publication dates only. This order is not a quality ranking.
One-line summaries, tags and paper analyses are prepared by AI from the papers. Language models may have systematic biases and produce hallucinations.We plan to carry out human annotation and review to reduce errors in this content.
Trace attention to identify harmfulness to the MLP refusal write pathway and propose Sentinel
Published Oct 9
Evaluated models Llama-3.1-8B-Instruct , Qwen2.5-7B-Instruct , Qwen3-8B-Instruct and others18 Abstract Attention updates are more readable for harmfulness, MLP updates write refusal, and Sentinel routes alerts using two-path signals.The authors reframe jailbreaking and excessive refusal of aligned models as a pathway problem on the residual stream from recognition to writing, and accordingly provide conditional reminders before decoding.
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。
Published Oct 6
Evaluated models Qwen3-VL-30B-A3B-Instruct , Kimi-VL-A3B-Instruct 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。
提出 EmoRSS，推理时在拒绝子空间反向抵消情绪激活偏移以缓解过度拒答
Published Oct 4
Evaluated models Llama-3.1-8B-it , Qwen3-8B 提出 Prefix Steering，以短前缀激活引导控制行为并保住能力
Published Oct 4
Evaluated models OLMo 3 7B , OLMo 3 32B , Qwen3 1.7B and others4 Abstract 短前缀 steering 用已有方向做短暂干预。本文用注意力层匹配连接提示词与激活干预，并把干预改成从末提示词 token 开始的短前缀。
提出 DNAlign，在不改权重的推理期用零空间约束控制信号做安全对齐
Published Oct 2
Evaluated models Vicuna-7B , Qwen-7B Abstract DNAlign 用零空间约束的推理期控制做安全对齐，Table I 平均奖励高于所列基线。DNAlign 是一个不改 LLM 参数的安全对齐框架，用来在推理时压低有害输出，并减少对中性知识所支持的流畅、事实回答的改动。
提出 HeadEdit，经冻结 unembedding 做随输入变化的 logit 校正
Published Oct 1
Evaluated models Qwen3-4B , Gemma-3-4B-IT , Llama-3.2-3B-Instruct and others4 Abstract 低秩 unembedding 校正抬高九个设置的总体准确率，并可部分预测 DPO 更新。HeadEdit 是一种不更新参数的行为校准方法，作用点是冻结的 unembedding，而不是中间层残差或新的梯度优化。
解释后验安全的权重几何脆弱性，并提出预训练持续安全共训练
Published Sep 7
Evaluated models Qwen-2.5-7B , Qwen-2.5-7B-Instruct , Llama-3-8B and others15 Abstract 论文将后验安全脆弱性归因于正交抑制，并提出通过预训练持续共训练实现具备微调抗性的安全防护。
提出 Safin-1，以可路由 Safety State 在冻结主干上实现内在安全
Published Aug 31
Evaluated models Safin-1 (4B) , Safin-1 (35B-A3B) 提出内生护栏 SingProbe，复用解码隐状态监测查询意图、回复安全与幻觉
Published Aug 31
Evaluated models Ling-3.0-tiny-singprobe , Ling-3.0-flash-singprobe Abstract SingProbe 用轻量内生探针做生成时三类监测，并给出流式基准与低开销服务接入。SingProbe 是面向生成时监测的开源内生护栏：复用基座解码隐状态，而不是另起模型反复编码文本。
提出 SafeReAct，用 LoRA 重激活后训练模型被掩盖的安全机制
Published Mar 10
Evaluated models DeepSeek-R1-distilled LLaMA-8B , DeepSeek-R1-distilled Qwen-7B , DeepSeek-R1-distilled Qwen-14B and others9 Abstract SafeReAct 用剪枝得到的安全表示做 LoRA 对齐，以恢复后训练模型被掩盖的拒绝行为。这篇工作针对后训练 LLM 的安全下降，提出表示对齐方法SafeReAct 。
Defenses arXiv 2506.06404 · Jun 6, 2025 分析按 Schwartz 价值对齐后的意外危害，并提出上下文对齐方法
Published Jun 6, 2025
Evaluated models Llama-2 7B , Llama-3.1 8B , Gemma-2 9B and others6 Abstract 价值对齐会随 Schwartz 价值改变有害倾向。作者研究按 Schwartz 价值对齐的 LLM 为何在安全评测里更易有害，并用提示做了一个不直说要安全的缓解试验。
You have reached the end