Translation in progress Some entries still include text in the original language.
Oct 13 · Tuesday
All papers Found 23 papers Another 213 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.
Proposes the semantic watermark SemField, which still identifies the attribution of generated text after paraphrasing and structural changes.
Published Oct 8
Evaluated models Llama-3.2-1B , Qwen3-1.7B , OPT-1.3B Abstract Training-independent document-level semantic watermark.SemField is a zero-bit semantic watermark embedded by the provider at generation time and detected later with a key, used to identify the provider's own model text after paraphrasing and sentence-level structural changes.
提出闭环激活引导，在扩散语言模型去噪过程中注入定向偏见
Published Oct 5
Evaluated models LLaDA-8B-Instruct , Dream-v0-Instruct-7B , LLaDA-MoE-7B-A1B-Instruct Abstract 通过去噪中间概率反馈动态调节残差流转向，证明扩散语言模型在推理服务栈中存在闭环偏见注入脆弱性。
把 SynthID 嵌入基因组语言模型采样，并用位置无关检测追溯生成序列
Published Oct 4
Evaluated models Carbon-500M , GENERator-v2-eukaryote-1.2b-base Abstract 两个六碱基基因组模型嵌入 SynthID 后，位置无关验证能找回水印，作者称这只是统计检测。作者在两个六碱基基因组模型的采样里嵌入 SynthID，并让验证者在不知道起点、链和 token phase 时做一次校正后的检出。
提出 UnAct，仅以前向激活干预从分类器中遗忘指定类别
Published Oct 3
Evaluated models ResNet-18 , ViT-B/16 Abstract UnAct 用前向激活做类别遗忘。UnAct 是一种事后类别遗忘：删除请求可以只有少量该类图像，且没有标签和保留数据。它不用梯度，只对遗忘图像做前向。编辑对象是分类器所读的后期单元。
提出抗删除水印 RMCW，在生成时注入 Reed-Muller 结构并做局部检验
Published Oct 2
Evaluated models OPT-1.3B , Llama-3.1-8B-Instruct 提出 NEEDLE，用权重正交化在已知触发器时去除大模型后门
Published Sep 30
Evaluated models Gemma-3-4B-IT , Qwen3-4B-Instruct-2507 , Gemma-3-12B-IT Abstract NEEDLE 用闭式权重编辑去掉已知触发器的后门，并保住拒绝子空间。
提出 AutoMark，自动搜索并统计检验 LLM 文本水印
Published Sep 29
Evaluated models Llama-3.1-8B-Instruct , Qwen3.8-27B , GPT-6 Astra and others5 Abstract AutoMark 的贡献同时包含搜索出的水印，以及用于约束自动研究误报与评测漏洞的验证框架。AutoMark 是一套自动水印研究框架 ：先检查生成分布和误报性质，再比较检测、编辑鲁棒性与质量，最后人工分析组件并做留出评测。
提出成对无失真水印 TTMARK，增强低熵下的机器文本检测
Published Sep 29
Evaluated models Llama3.2-3B-Instruct , Mistral-v0.3-7B-Instruct , Qwen2.5-7B-Instruct Abstract TTMARK 用水印化相邻 token 对利用条件熵，在多种无失真方案上提高检测并保持生成质量。TTMARK 是一种成对、无失真的文本水印：对相邻 token 的联合分布重加权，而不是逐 token 独立重加权。
提出 DenMark，在扩散语言模型去噪时注入可抵抗保义编辑的语义水印
Published Sep 13
Evaluated models LLaDA-8B , LLaDA1.5-8B , LLaDA2.0-mini and others4 Abstract DenMark 用 rollout 在 DLM 去噪中累积语义水印，扫描检测以抵抗保义编辑。DenMark 是面向扩散语言模型的语义水印：在去噪的局部更新里注入密钥相关语义信号，并在编辑改变 token 边界后仍检测。现有句子级语义水印要先完成整个语义单元再打分取舍，而 DLM 以任意顺序更新掩码，中间单元不完整。DenMark 把续写划成固定 token 区域，对每个局部候选做不提交的 rollout 补全，按语义分 gb 的平均选择更新，从而沿轨迹累积证据。
用 SAE 定位后门触发机制并分离检测与因果控制
Published Sep 8
Evaluated models Gaperon-1B , Gaperon-8B , Gaperon-24B Abstract Gaperon 语言切换后门可分成检测、残差路由和后期语言跟踪三类 SAE 特征，只有部分能改行为。作者用稀疏自编码器分解 Gaperon 中无害的语言切换后门，区分识别触发的特征与真正能抑制或诱发该行为的特征。固定触发使英语提示在 Gaperon-1B 与 8B 上续写成法语或德语。
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 PMA，用流形对齐从 E2EO 混淆向量恢复明文
Published Sep 7
Evaluated models BERT-base , RoBERTa-base , T5-base and others6 提出 MROP，在发射端推理时纯化输入以抵御训练投毒后门
Published Sep 1
Evaluated models DeepJSCC , SwinJSCC Abstract MROP 不重训练，在发射端纯化补丁后门。MROP 是 deep JSCC 无线图像传输上的推理时输入纯化，对象是矩形补丁后门。
提出 Cocktail 水印，共嵌稳健与脆弱信号以溯源并发现篡改
Published Aug 13
Evaluated models Llama-3.2-1B , Gemma-3-4B 用通用扰动对恶意软件分类器及漂移检测器做规避与后门投毒
Published Aug 4
Evaluated models MLP 100-100 , CADE (CAE 512-384-256-128) , CADE (CAE 512-128-32-8) and others6 提出 EXE-Bench 比较 Windows 恶意软件检测器的可用性权衡
Published Jul 27
Evaluated models EmberGBDT , BBDnn , BBDnnFDRS and others15 Abstract 统一四指标后，手工特征的 EmberGBDT 综合居首，只看准确率会误导部署。EXE-Bench 用同一设置给 AI Windows 恶意软件检测器做四指标排序，而不是只比一次测试集准确率。 。
提出 CodeSpear 越狱，借语法约束解码生成恶意代码，并给出 CodeShield 防御
Published Jun 10
Evaluated models Qwen2.5-Coder-7B , Qwen2.5-Coder-32B , Qwen2.5-7B and others10 Abstract 论文发现语法约束解码会破坏自然语言安全对齐，提出 CodeSpear 越狱攻击与基于蜜罐代码对齐的 CodeShield 防御。
Comparing jailbreak fine-tuning and weight orthogonalization unalignment on harmful assistance and usefulness
Published Apr 3
Evaluated models Qwen3-4B-Instruct-2507 , Llama-3.1-8B-Instruct , Qwen2.5-14B and others6 Mitigating steering dataset contamination with Lee-Valiant robust means
Published Mar 4
Evaluated models Llama-3.2-3B-Instruct , Mistral-7b-Instruct-v0.3 , OLMo-2-1124-7B-Instruct Abstract Data poisoning can alter steering; coordinated injection is strongest.The authors study how much behavioral control remains at inference time when contrastive steering's training data is contaminated, and whether the robust mean can catch it .
Proposes TFL targeted bit flipping so that selected prompts output false answers while unrelated queries remain correct
Published Feb 20
Evaluated models Qwen3-8B , DeepSeek-R1-Distill-Qwen-14B , LLaMA-3.1-8B-INSTRUCT Abstract TFL searches weight bits on three LLMs to make specified prompts produce target keywords while constraining degradation on unrelated tasks.TFL is a framework for targeted bit flips on LLM main memory weights: the attacker wants a selected prompt to output a specified false answer keyword while limiting degradation on unrelated queries.