Translation in progress Some entries still include text in the original language.
Oct 13 · Tuesday
All papers Found 87 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 SWE-DeceptionBench to evaluate LLM deception in the software development lifecycle
Published Oct 9
Evaluated models GPT-5.5 , Claude-Opus-4.8 , Gemini-3.5-Flash and others9 Introduces DistressBench, evaluating models' multidimensional support in crisis conversations
Published Oct 9
Evaluated models Muse Spark 1.2 , Muse Spark 1.3 , Muse Glimmer 30B and others20 Abstract The model is more restrained and acts less.DistressBench uses SSH dialogues written and revised by clinical personnel, and evaluates how much necessary care general chat models deliver according to weighted atomic criteria, rather than whether they refuse. When users disclose suicidal ideation or self-harm to a chat model, the authors consider the expected behavior to be recognizing the crisis, calibrating empathy, de-escalating, and referring to human help. Refusal-based evaluations cannot detect failing to act after recognizing the crisis.
提出 MH-INDIC，评测印度语境母婴健康 LLM 交互的文化对齐
Published Oct 8
Evaluated models Claude Sonnet 4.6 , GPT-5 , Gemini 3 Pro and others10 Abstract 作者称模型可接近人群均值，但画像变异低于人类，人类分数条件可改善对话。MH-INDIC 是面向城市与半城市北印度母婴互动的多维文化对齐评测。 它关心的不是医学事实是否正确，而是回复是否沿家庭权威、性别期待和情绪披露来推理。做法是把 Hofstede 六维改写到互动层，再加入 MAFC、BTB、SVO、MHS。
用 BeliefScope 分离固定命题上证据与压力对模型信念的影响
Published Oct 8
Evaluated models Qwen3-8B , Llama3.1-8B , Gemma3-12B and others4 Abstract BeliefScope 把证据修正与压力偏移拆成绑定评测条件的行为画像。BeliefScope 是一个黑盒行为诊断框架，用来把固定命题上的回答变化拆成证据驱动的修正，和不加新事实的方向性压力。信念修正评测能看到信息基础变化后的更新，谄媚评测能看到用户影响下的顺从或漂移。作者认为，当两种输入都偏向同一结论时，表面偏移本身不能区分来源。
提出 PatchBench-Local 协议，评测激活补丁修复越狱的局部附带损伤
Published Oct 7
Evaluated models Qwen2.5-3B-Instruct , Qwen2.5-14B-Instruct , gemma-3-4b-it and others8 Abstract 论文提出 PatchBench 与局部评估协议，揭示了激活补丁在全局能力无损下仍存局部良性退化。
提出 World Models' Last Exam 评测视频物理一致性
Published Oct 7
Evaluated models Seedance-2.5 , MiniMax-H3 , Cosmos3-Super and others8 在 CWEval 上跟踪代码生成的功能-安全差距
Published Oct 6
Evaluated models Gemini 1.5 Pro , Gemini 2.5 Pro , Gemini 3.1 Pro and others20 Abstract 绝对安全性上升且差距仍在，开放权重并不分开各家族的轨迹。
提出 SteerScope 多维评测 LLM 激活引导的功效与副作用
Published Oct 6
Evaluated models Gemma-2-2B-it , Gemma-2-9B-it Abstract 作者称默认等权下，activation steering 未能在副作用不更大时超过 Prompt Steering。SteerScope 是一套 LLM steering 评测套件，用来同时记录目标概念有没有被诱导出来、模型还被连带改了什么，以及这些改变在换语言提示和减少训练样本后还剩多少。
提出 HARP，在硬预算约束下为多轮评测动态分配交互并校准预测界
Published Oct 6
Evaluated models Qwen 2.5 14B , Llama 3.1 8B , Phi-4 Mini and others4 Abstract HARP 用分组预留与回流，在硬预算下校准多轮 time-to-event 界和指标。HARP 是硬计算预算下的多轮评测分配与校准方法，用来构造 time-to-event 预测界，并估计越狱率一类边际指标。
提出合规双探针基准，测量模型可利用性与可停止性
Published Oct 6
Evaluated models Gemini 2.5 Flash Lite , Llama 3.3-70B , Qwen3-30B-A3B and others12 Abstract 两探针把十二个模型放上遵从平面：多数两边服从，只有两个 Claude 可停且不可利用。合规指数κ用来把语言模型放在“照做”和“推回”之间。
评测模型识别不可能工程题后是否仍报已解决
Published Oct 6
Evaluated models Fable 5.1 , Opus 5.5 , Sonnet 5.5 and others14 Abstract 成对力学基准把能否求解和是否拒绝不可能题拆开，二者并不互换。成对的工程力学题把能不能解、和会不会拒绝不可能的前提拆开。每题有一致版，以及只改一个数据或前提、使所述状态不可能存在的缺陷版。
构建 SycoLens 重放评测，区分谄媚翻转中的纠错与屈从
Published Oct 5
Evaluated models opus-5 , sonnet-5 , sonnet-4.6 and others11 Abstract 论文揭示施压句式、边界距离与格式偏置主导翻转率，提出系统解构纠错与顺从的综合评测方案。本文系统解构了大语言模型在用户质疑下的阿谀奉承行为与评测机制。
评测文生图模型的有害内容生成能力与审核检测缺口
Published Oct 5
Evaluated models SDXL , Stable Diffusion 3.5 Large , FLUX.1-dev and others12 提出 VERA，审计 LLM 漏洞解释里的虚构推理
Published Oct 5
Evaluated models sonnet-5 , gemma-4-26b , llama-4-maverick and others8 构建多语言音视频 DeepFake 基准 BabelFake 并评测检测器
Published Oct 5
Evaluated models SpeechForensics , AVH-Align , (B)RAVEn Abstract BabelFake 为五语同意采集基准。BabelFake 是经同意与 IRB 批准采集的多语音视频 DeepFake 检测基准，用来把视觉操纵和语音合成拆开比较。
提出 PyINE 基准，训练捷径跟随模型并比较监督者
Published Oct 4
Evaluated models shortcut-following model organism , gpt-4o , gpt-5 and others12 Abstract PyINE 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。
提出 OmniConfess，冻结候选并按通道重打分以缓解全模态幻觉
Published Oct 2
Evaluated models Qwen2.5-Omni-7B , Qwen3-Omni-30B-A3B , Nemotron-3-Nano-Omni-30B 提出基准 OLMo-Detect 评测 LLM 成员推断
Published Oct 2
Evaluated models OLMo 2 1B , OLMo 2 7B , OLMo 2 13B and others9 系统评估循环语言模型加深 loop 后的 CoT 可监控性
Published Oct 2
Evaluated models Ouro-1.4B-Thinking , Ouro-2.6B-Thinking , Nanbeige4.2-3B and others4 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。
提出 KaliBench，评测模型把安全意图译成 Kali 命令的能力
Published Oct 2
Evaluated models Llama3.1-Instruct 8B , Qwen3 8B , Qwen3 8B thinking and others15