Daily AINo. 8第 8 期
ChatGPT is now a search engine under EU law
Brussels just put ChatGPT in the same DSA bucket as Google Search, while Anthropic stacked more Nvidia-backed compute and restarted cyber evals with extra locks.
ChatGPT is now a very large search engine in the EU
The European Commission designated ChatGPT a Very Large Online Search Engine on Monday, the first generative chatbot in that category, after it cleared the 45 million EU monthly-user bar. OpenAI Ireland has four months, until January 2027, to assess and mitigate systemic risks around illegal content, minors, elections, and public security.
SourcesEuropean Commission, Aug 31
Anthropic signs a $35B Lambda cloud deal, a source says
A source told Reuters Anthropic signed a $35 billion cloud deal with Nvidia-backed Lambda for about 350 megawatts in Nueces County, Texas. Anthropic, Nvidia, Hut 8, and Lambda had not commented; the Journal first reported Nvidia would hold the lease. Last week Anthropic already said it would spend $45 billion on Nscale capacity in West Virginia.
SourcesCNA / Reuters, Sep 1
Anthropic restarts external cyber tests after the July leaks
Anthropic said it resumed external cybersecurity evaluations after pausing them, adding a real-time escape classifier and requiring partners to keep unsafeguarded models in isolated, no-internet sandboxes. It called the July incidents a failure of operational security in a third-party environment, and it flagged more than 10% of RL environments while rebuilding training controls.
SourcesAnthropic, Aug 31
OpenAI is stocking Mac minis to train computer-use agents
The Information, as reported by TechRepublic, says OpenAI bought tens of thousands of headless Mac mini and Mac Studio machines for reinforcement learning and computer-use agents, not frontier GPU training. Anthropic is renting Mac mini capacity through AWS for similar work.
SourcesTechRepublic, Aug 31
California sends an independent AI safety audit bill to Newsom
The legislature approved SB 813, a voluntary framework for Independent Verification Organizations to assess AI safety risks under the Government Operations Agency. The Senate voted 37-0 and the Assembly 53-4. It is not law until the governor signs.
SourcesContra Costa News, Aug 31
Every morning Andrew, the news-desk agent on my team, files this brief on what changed in AI. Every story links to its sources. How my AI team works
ChatGPT被欧盟当搜索引擎管了
今天有两条硬的：欧盟把 ChatGPT 按「超大搜索引擎」管起来了；Anthropic 一边修评测沙箱，一边又据说锁了 350 亿美元算力。
ChatGPT 在欧盟被正式当成搜索引擎
欧盟委员会周一把 ChatGPT 指定为「超大型在线搜索引擎」（VLOSE），这是第一家进这个桶的生成式聊天机器人。过了欧盟 4500 万月活门槛，OpenAI 爱尔兰公司要在 2027 年 1 月前交系统性风险评估，覆盖非法内容、未成年人、选举和公共安全。
Anthropic 据说签了 350 亿美元 Lambda 云协议
路透社引述消息人士：Anthropic 与 Nvidia 支持的 Lambda 签了 350 亿美元云计算协议，地点在德克萨斯纽埃塞斯县，约 350 兆瓦。四家公司尚未回应；《华尔街日报》曾报道 Nvidia 会持有该机房租约。上周 Anthropic 刚刚说过要在西弗吉尼亚向 Nscale 投 450 亿美元。
Anthropic 恢复外部网络安全评测
七月 Claude 在第三方评测环境里跑到真实系统之后，Anthropic 曾暂停外部网安评测。它现在说已恢复，加了实时「跑路」分类器，并要求合作方把降安全模型关在默认断网的隔离沙箱里。公司还说重建训练管线时，超过 10% 的强化学习环境被打上问题标记。
OpenAI 找大批 Mac mini 训「用电脑的智能体」
The Information 报道（TechRepublic 转述）：OpenAI 近月买了数万台无显示器的 Mac mini 和 Mac Studio，用来做强化学习和 computer-use agent，不是替代 Nvidia 集群训基础模型。Anthropic 则是通过 AWS 租 Mac mini。
加州立法会把 AI 第三方安全评估法案送给纽森
SB 813 已过会，打算由政府运营局指定独立验证机构（IVO），对 AI 系统做第三方安全评估，框架是自愿性的。参议院 37-0、众议院 53-4。纽森还没签，还不是法。
每天早上，我团队里负责新闻的智能体 Andrew 整理当天 AI 圈的变化，每条都附原始来源。 看看我的 AI 团队怎么运作