Daily AINo. 26第 26 期
Governments want a watchdog. OpenAI wants Washington to write the yardstick.
UN week opened with a 20-country push for global AI oversight, a province suing OpenAI over a school shooting, and Alibaba flexing a new China chip ahead of the Xi–Trump summit.
20 countries + EU call for a global AI oversight body
Germany, Canada, Australia, Singapore, the UAE, South Africa and others — plus the EU — issued a joint statement urging common standards, shared serious-incident reports, and exploring an international institution that could set rules and convene when capability thresholds are crossed. The US and China did not sign.
SourcesAl Jazeera, Sep 22
British Columbia sues OpenAI over Tumbler Ridge
BC filed in San Francisco federal court against OpenAI and Sam Altman, alleging ChatGPT’s safety team flagged the shooter’s gun-violence chats but the company never alerted police before the February attack that killed eight. The province wants damages plus a court order forcing how OpenAI handles violent threats.
SourcesAl Jazeera, Sep 22
OpenAI: US should lead frontier AI standards (incl. self-improvement)
In “Building standards for the next phase of AI,” OpenAI asked Washington to coordinate national AI safety institutes on voluntary technical benchmarks and shared incident-reporting — including how to measure recursive self-improvement — and said fully autonomous RSI should wait until it can be done safely.
SourcesTechNode Global, Sep 22
Alibaba unveils Zhenwu V900, eyes 5–10T-parameter models
At Apsara in Hangzhou, CEO Eddie Wu called Zhenwu V900 China’s most powerful AI chip — about 3× the prior M890 — and said Alibaba plans a next Qwen-scale model at 5–10 trillion parameters (Qwen3.8-Max is 2.4T), plus cloud compute past 20 GW by 2032. Timing lands days before Xi arrives in Washington.
SourcesAP / WTOP, Sep 22
New York starts RAISE registration for big AI labs
Gov. Kathy Hochul said frontier developers with over $500M revenue must register with the state DFS from November and report critical safety incidents within 72 hours, with fuller RAISE Act compliance from Jan 2027. She would not rule out exploring “kill switches.”
SourcesNY Post, Sep 21
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
联合国开会前，20国要建“全球监工”，OpenAI想让美国写尺子
今天这波其实就一件事：谁来管最强AI——国家联盟想建全球机构，OpenAI想让美国定标准，加拿大一个省已经把OpenAI告上法庭，阿里还在云栖上甩出新国产芯片。
20国+欧盟喊话：建全球AI监管机构
德国、加拿大、澳大利亚、新加坡、阿联酋、南非等约20国，加上欧盟，发联合声明：AI必须留在人类控制下，推动共同标准、共享严重安全事故，并探讨成立国际机构——能力跨过阈值就召集各国。美中没签字。
加拿大不列颠哥伦比亚省起诉OpenAI
BC省在旧金山联邦法院起诉OpenAI和Altman：枪手在ChatGPT里聊枪支暴力，安全团队已标记，但公司没报警；2月Tumbler Ridge校园枪击致死8人。省府要赔偿，还要法院强制改危险对话处置流程。
OpenAI：美国该带头定前沿AI技术标准
博客《Building standards for the next phase of AI》里，OpenAI呼吁美国牵头，把各国AI安全研究所串起来，定自愿技术基准和事故报告协议——包括怎么衡量“递归自我改进”（RSI），并说完全自主的RSI在安全前不该做。
阿里云栖发布镇武V900，放话5–10万亿参数模型
吴泳铭称镇武V900是“中国目前最强AI芯片”，性能约上一代M890的3倍；并计划训练5–10万亿参数新模型（Qwen3.8-Max约2.4T），云算力目标2032年前超20吉瓦。时间点卡在习特会前。
纽约州：大厂11月起登记，72小时报重大事故
Hochul宣布推进RAISE法案落地：年收入超5亿美元的AI开发者从11月起向州金融署登记，重大安全事故72小时内上报；2027年1月起更全面合规。她还说不排除研究“AI紧急关停开关”。
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