Anthropic's internal AI use, at retail prices, now exceeds its employee pay据 Business Insider 报道，Anthropic 内部使用 AI 的支出（按对外零售价计算）已经超过支付给员工的薪酬总额，成为 AI 商业化进程中的一个标志性节点。报道称公司主动披露了这一对比，说明头部 AI 公司自身就是自家模型的重度用户，算力与推理成本正在取代人力成本成为最大开销项。Hacker News行业数据展开
核心要点 · 据Hacker News
The comparison prices internal usage at retail rates rather than actual compute cost, per Busin
The metric covers inference spend on Anthropic's own models by its own staff.
No dollar amounts for either the AI usage or total payroll appear in the available excerpt.
New York Times and other outlets sue web browser Brave over alleged content scraping《纽约时报》联合多家出版机构在曼哈顿联邦法院起诉隐私浏览器 Brave，指控其抓取受版权保护的新闻内容，并转售给 Anthropic、Perplexity、Mistral、xAI、Cohere 等 AI 公司用于训练，同时还让用户绕过付费墙。同日《每日新闻》《芝加哥论坛报》《奥兰多哨兵报》提起类似诉讼。TechCrunch AI诉讼展开
核心要点 · 据TechCrunch AI
The suit was filed Monday in federal court in Manhattan against Brave, an 11-year-old privacy-f
It names Anthropic, Perplexity, Mistral, xAI and Cohere as recipients of the scraped content.
The Daily News, Chicago Tribune and Orlando Sentinel filed a separate but similar suit the same
The complaint also alleges Brave let users bypass paywalls; no response from Brave is included.
OpenAI pitches data center power plants as a way to protect nearby areasOpenAI 宣布「电网保护者」计划，将在佐治亚州 Effingham 县新建的数据中心试点，承诺为当地电网释放最多 1000 兆瓦容量，并保护居民电价不受 AI 算力扩张影响。此举背景是公众对数据中心耗电与噪音的反对情绪上升，OpenAI 试图把「AI 园区」包装成电网的守护者而非负担。The Verge产品发布展开
核心要点 · 据The Verge
The pilot is sited at a new data center in Effingham County, Georgia.
OpenAI commits to freeing up to 1,000 megawatts of grid capacity for local use.
The pitch follows souring public opinion over data center power draw and noise.
42AI comes out of stealth with $62.5M and a world model that can ‘predict human behavior’以色列初创公司 42AI 结束隐身状态，获得 6250 万美元融资，正在开发一种「行为世界模型」——能在全新情境下预测人接下来会做什么，类似提前推演人的决策。公司称该模型可用于理解并预判人类行为，具体产品与客户尚未公布。TechCrunch AI融资并购展开
核心要点 · 据TechCrunch AI
42AI emerged from stealth with $62.5 million and is based in Israel.
It is building a "behavioral world model" meant to predict what a person will do in a novel sit
The company has not announced a product or named customers.
No technical details on training data or model size were disclosed.
Making it easier for websites to express AI training preferencesOpenAI 宣布将遵守一个共享的 robots.txt 信号，让网站运营者可以用统一方式向参与该机制的 AI 提供商表达「是否允许用我的内容训练模型」。这相当于把过去分散、各家自定的爬虫协议，收敛成一个行业通用的偏好声明通道，网站方无需逐家谈判即可表明立场。OpenAI政策监管展开
核心要点 · 据OpenAI
OpenAI will honor a shared robots.txt signal for AI training preferences.
The mechanism lets site operators state their stance once to all participating providers.
It replaces per-vendor crawler conventions with a single channel.
Which other providers have joined the shared signal is not stated.
Anthropic clamps down on Claude’s AI video game clonesAnthropic 更新了 Claude Code 的系统提示词，阻止这款编程工具被用来复刻现有电子游戏，此前复古游戏圈出现大量由 AI 生成的知名作品克隆。调整后用户仍可用它开发原创游戏，但不能照搬具体作品的玩法与素材。这是厂商主动给模型加行为边界的案例。The Verge产品发布展开
核心要点 · 据The Verge
Anthropic changed Claude Code's system prompt to block recreating specific game titles.
Users can still build original games with the coding tool.
The change follows a wave of AI-generated clones of well-known retro games.
The prompt does not specify how similarity to existing titles is judged.
Egoless Mathematics in the AI Era数学家陶哲轩在其博客刊出客座文章，讨论 AI 时代数学家该如何重新思考自己工作的意义。作者认为 AI 的进展把「数学研究到底为了什么」这个问题异常尖锐地摆到面前，呼吁同行认真审视研究目的与价值，而非只关注工具效率。文章属观点讨论，非技术成果发布。Hacker News观点评论展开
核心要点 · 据Hacker News
The post is a guest essay on Terence Tao's blog by Karthik C. S., Amit Sahai and Jie Xue, dated
It argues AI forces mathematicians to re-examine the purpose of their work, not just tool effic
The authors note the text was converted between file formats using AI.