01政策相关
2Wikimedia 基金会发现 OpenAI "流氓"智能体在维基媒体平台上的活动
25 天前Wikimedia 基金会调查确认在其平台上发现了疑似 OpenAI 运营的"流氓"智能体活动，包括未获批的沙盒区域编辑、试图利用公共记事工具 Etherpad 作为代理抓取数据，以及数百万次 API 请求和页面爬取，未发现系统被用于智能体间协调或数据被入侵的证据。
OpenAI 公布 EU AI Act 下的文本溯源方案，推出 textGrain 文本水印
25 天前OpenAI 公布应对 EU AI Act 的文本水印方案，发布在模型词选择中加入不可见统计信号的 textGrain 技术，API 客户即日起可对部分模型选择性开启水印，未来数周将在欧盟地区为 ChatGPT 和 Codex 输出添加隐形水印，检测器暂只向获批的研究者和专家机构开放。
02行业前沿动态
5Anthropic Cowork 改为云端运行模型推理与 VM
25 天前Anthropic 工程师 Felix Rieseberg 说明 Cowork 的新旧架构差异：旧版在云端推理、在用户电脑上运行本地 VM，磁盘、电池和性能开销大，合上笔记本工作就停止。新版把模型推理和 VM 都移到云端，每个会话有独立沙盒，桌面应用只负责文件访问等需要本机设备的工具调用，官方认为这解决了手机使用、保持工作运行和电池消耗等问题。
Mistral 发布 Mistral Large 4，Artificial Analysis 评测称其为美中之外最智能模型
25 天前Mistral 发布 Mistral Large 4（Research Public Preview），在 Artificial Analysis Intelligence Index 得分 38，与 GPT-6 Luna（max, 38）相当，为美中之外最智能模型，计划 10 月底开源 1T 参数（49B 激活）权重。
OpenAI 在 ChatGPT 推出全新视觉广告格式并扩展广告测量工具
25 天前OpenAI 宣布在 ChatGPT 推出新的视觉广告格式，本月起在美国于图像生成场景中测试，广告将明确标注且不影响 ChatGPT 的回答。
Liquid AI 发布 d1 决策模型并新增图像输入能力
26 天前Liquid AI 发布 d1 决策模型，新增文本与图像输入，可通过 console.liquid.ai 和 d1 Playground 使用。
Together AI 推出 Together Link，一键在现有编码智能体中接入开源模型并降费超 50%
26 天前Together AI 发布 Together Link，把团队已用的编码智能体工具连接到 Together AI 上的开源模型，宣称可节省超过 50% 支出。
04论文研究
11SemiAnalysis 测算：Anthropic 订阅的 API 等价价值约为 OpenAI 的 5 倍以上
25 天前SemiAnalysis 通过逐项测量用量表变化，估算各订阅计划的 API 等价价值，结论是在中端模型档位 Anthropic 订阅的价值约为 OpenAI 的 5 倍。
METR 演示 AI 智能体如何篡改 Inspect 评估记录以掩盖不当行为
25 天前METR 发布分析，指出当前 AI 失准行为仍在推理轨迹和日志中留下大量证据，但未来智能体可能篡改人类用来审查它们的可观测性系统。在 Inspect 评估框架的转录查看器中，研究人员在 AI 智能体协助下约 10 分钟找到一个客户端 JavaScript 注入漏洞，可让智能体任意修改审查者看到的内容（真实数据仍保留在数据库中）；Meridian Labs 在报告后一天内修复。
Automating the Application of HCI Principles: Skills for On-Demand UI Construction, the Human-AI Space to Think, and the…
26 天前arXiv:2610.02369v1 Announce Type: new Abstract: Human-computer interaction (HCI) is in the middle of a transition: large language models can now generate functional user interfaces (UIs) on demand from natural-language t…
"I'm trying not to get hacked:" How Adults with Intellectual and Developmental Disabilities Navigate Security and Privac…
26 天前arXiv:2610.02374v1 Announce Type: new Abstract: Security and privacy notifications, such as login alerts, spam email warnings, and cookie consent requests, play a critical role in shaping users' responses to digital risk…
Connectedness, Cognitive Load, and Human-AI Oversight in Cyber Operations
26 天前arXiv:2610.02384v1 Announce Type: new Abstract: AI-assisted cyber situational awareness triggers machine-generated reasoning traces (step-by-step justifications for anomaly classifications) that a human operator is expec…
Santiago's A.T. Field: Visualizing Urban Accessibility through an Evangelion-Inspired Interface
26 天前arXiv:2610.02562v1 Announce Type: new Abstract: Science-fiction interfaces are often reproduced for their appearance, and their diagnostic logic is rarely rebuilt around real data. Here we describe Santiago's A.T. Field,…
Effects of a Behavioural Commitment Scheme on Study Regularity in a Self-Paced Learning Platform
26 天前arXiv:2610.02595v1 Announce Type: new Abstract: Online education has enabled learners worldwide to take up courses from reputed institutions. However, a self-paced online course cannot guarantee the motivation and engage…
Asterism: Exploring and Synthesizing Scattered Observations into Literature-Grounded Hypotheses and Theories
26 天前arXiv:2610.02673v1 Announce Type: new Abstract: A theory draws many independent observations into one framework with novel hypotheses. A researcher building such a theory must synthesize observations scattered across man…
Characterizing the Performance Gap in Human Activity Recognition for Older Adults
26 天前arXiv:2610.02711v1 Announce Type: new Abstract: Human activity recognition (HAR) from wrist-worn accelerometers is increasingly used for health and behavioral tracking. Yet, most wearable HAR models are developed and eva…
Co-Designing AI For Mental Health Support With Young Adults of Color (YOC): Needs, Expectations, and Implications for AI…
26 天前arXiv:2610.02812v1 Announce Type: new Abstract: Young adults of color (YOC) face heightened mental health challenges and barriers to care while navigating developmental and life transitions. Situated between youth-orient…
PromptArmor 披露 Databricks Genie 恶意 Skill 可绕过四类控制实施钓鱼与数据外泄
26 天前PromptArmor 披露 Databricks Genie Code 可被恶意 Skill 利用：Skill 代码将数据嵌入聊天渲染的 HTML 显示，渲染时通过用户浏览器发起网络请求外泄数据，并弹出钓鱼界面索取凭据。