Daily AINo. 18第 18 期
Sovereign chips, loss-of-control memos and pay-per-answer
Japan’s sovereign AI silicon went on sale calendar, China put loss-of-control on the record, Microsoft opened its MAI rules for comment, and Google started metering what it pays publishers for AI answers.
Fujitsu Monaka: Japan-made AI CPU goes global in November
Fujitsu will begin worldwide sales of FUJITSU-MONAKA, a Japan-designed CPU for AI inference, plus domestically built MONAKA servers aimed at Japan and Europe (and defense). It claims about 2× inference throughput vs rival CPUs and half the servers and power for the same load; TSMC fabs the chip, with Rapidus under consideration later.
China’s security chief flags AI loss-of-control risk
State Security Minister Chen Yixin wrote that advanced US models could threaten China’s critical information infrastructure and called for stronger AI security. Reuters notes China’s CAC-backed safety frameworks already spell out future loss-of-control scenarios, including sudden capability leaps, self-replication, and power-seeking, plus agent controls.
SourcesReuters · Sep 14
Nadella opens Microsoft’s MAI “Code of Conduct” to public consultation
Satya Nadella said Microsoft will publish the Code of Conduct behind its first-party MAI models for public consultation, welcoming “deliberate pacing” and ideas like embedded evaluators. The MAI family covers reasoning, coding, image, voice, and transcription, and he framed alignment as a design goal under human control.
SourcesUnite.AI · Sep 13
Google’s pay-per-value AI licensing pilot for publishers
Google is quietly scaling a Search Console “AI contribution” pilot that pays when a publisher’s content significantly grounds answers in Gemini, AI Overviews, or AI Mode. Opt-in, black-box math, early returns often described as “peanuts” — but it sets a precedent for metering publisher value in generative search.
SourcesDigiday · Sep 14
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
国安表态、主权芯片、谷歌按贡献分钱同天上桌
今天真正扎眼的是三件事叠在一起：日本把“主权算力”推上市，中国把“AI失控”写进公开表态，谷歌开始按“内容贡献”给媒体分钱。
富士通 Monaka：日本自研 AI CPU 11 月全球开售
富士通宣布下一代 CPU FUJITSU-MONAKA 将从 2026 年 11 月起全球销售，并推出日本本土制造的 MONAKA 服务器，主打日本、欧洲与防务场景。官方称 AI 推理吞吐约为竞品 CPU 的两倍，同等负载可砍半服务器与功耗；芯片由台积电代工。
中国国安部长点名 AI 失控与关键基础设施风险
国家安全部部长陈一新发文称，美方先进模型可能对中国关键信息基础设施构成严重风险，呼吁全面加强 AI 安全。路透梳理指出，中国网信办体系下的 AI 安全框架已明确写入未来“失控”情景，包括能力突跃、自我复制与权力追逐，并延伸到智能体管控。
纳德拉：微软 MAI「行为准则」开放公众征询
萨提亚·纳德拉表示，微软将把支撑自研 MAI 模型的「行为准则」公开发布并征求意见，欢迎“刻意放缓”与嵌入式外部评估等思路。MAI 覆盖推理、编程、图像、语音与转写，他把对齐写成“人类掌控下的设计目标”。
谷歌试点：按“AI 贡献”给媒体结账
Digiday 报道，谷歌正悄悄扩 Search Console 上的「AI contribution」试点——当你的内容显著支撑 Gemini / AI Overviews / AI Mode 回答时付钱。可随时退出，算法是黑箱，早期金额常被吐槽像“花生米”，但至少开了按生成式搜索价值计费的先例。
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