SkillHub.news SkillHub.news 月报

MONTH 2026-09 · 52 STORIES · SKILLHUB MONTHLY

SkillHub.news 每月AI

2026-09 · 月报 · 素材 4

本期看点52 条 · 约21分钟

本月主题线MONTHLY THREADS24 条

费半-5.86%：降速呼吁首个定价日

官方事件精选OFFICIAL EVENTS20 条

Detecting wildfires with ChatGPT

榜单异动RANKING MOVES8 条

obra/superpowers 星标 +101,959

01本月主题线MONTHLY THREADS24 条

费半-5.86%：降速呼吁首个定价日

多源2026-09-15置信度高

9/14费半-5.86%（7/1来最大单日跌幅），英伟达-3.36%/美光-5.25%/英特尔-5.59%/康宁-13.7%；纳指仅-0.56%，CrowdStrike+13.85%创新高、软件股涨4-7%——同日完成硬件→软件/安全轮动，AI行情叙事切换首次被市场确认(多源,高)

美参议院"注意义务"法案草案 vs 特朗普否决

Reuters2026-09-15置信度高

图恩/克鲁兹/克洛布彻讨论：要求AI公司自证防伤害+授权商务部长派审查员实测产品，仍为草案成法概率低；特朗普9/14暗示不签——联邦"立法vs行政"分裂定型，与9/16 Ban ASI Act吹风会同周共振(Reuters,高)

国产开源三榜齐升（GLM-5.3+DeepSeek V4.1-Flash）

SuperCLUE/AA官网2026-09-15置信度高

SuperCLUE 9月期（官网9/11批次）：GLM-5.3 71.29（+8.02）第5跳第3、V4.1-Flash 71.81首发即国产第2；AA智能指数GLM-5.3 44.9开源第一（K3 43.8/Flash 39.5）；LMArena公司榜GLM-5.3第5——三口径互证，后训练+架构换新非预训练突破(SuperCLUE/AA官网,高)

OpenRouter新周127万亿：中国连续20周超美

OpenRouter/每经2026-09-15置信度高

9/7-9/13全球126.8-127万亿Token（环比+10%）、API调用55.5亿次；中国61.17万亿vs美国21.76万亿；小米MiMo-V2.5 +230%进前五；V4.1 Flash上线3天列第6；官方榜与聚合榜统计窗口不同排名不一致，引用须注明窗口(OpenRouter/每经,高)

宏观三线收紧：10Y首破5%+管道遇袭+FOMC 90-92%（进展·第1天）

新华社/CME2026-09-15置信度高进展·第 1 天

10Y盘中5.017%（2023年10月来首破）；沙特输油管道遇袭受损停运（非主动关闭），布油收105.68/WTI 101.39（盘中109.8/104.95）；CME加息概率升至90-92%（9/14早间86.2%）；今夜FOMC+8月零售销售同日落地(新华社/CME,高)

OpenAI预训练安全协议落地（进展·第3天）

OpenAI官方2026-09-15置信度高进展·第 3 天

官方博客+Altman X：前沿RL run启动前须制定safety cases（安全案例门槛）——"Dario长文→Altman附议"的口头共识首次固化为协议文本，头部实验室版本排期多一道安全闸门(OpenAI官方,高)

九万里未来科技超1亿美元种子+天使轮

9/14多方报道2026-09-15置信度中高

前地平线芯片总裁陈鹏创立，边侧/端侧大算力Agentic推理芯片；红杉中国/华业天成/蓝驰/联想创投/中科创星/武岳峰参投、高鹄独家FA——一级市场钱继续直插芯片/算力底座(9/14多方报道,中高)

券商智能体采购爆发+730天订阅制

报道台账2026-09-15

长江证券"金融对话智能体云服务"招标（9/8公告、9/14报名截止、9/24投标截止）；国投证券AI智能工作台授权服务协议730天（9/3公告）；8月9家券商10条需求渗透合规/交易/投研；银行扩散：重庆农商行两标段超1000万——AI采购进入同业复制期(公告/智能超参数,高/中高)

小红书开源Search Agent Iris+书生-S2科研基座

报道台账2026-09-15

Iris-mini 35B/Iris-pro 397B（基于Qwen，BrowseComp 82.2/88.6同量级开源最强，权重+评测代码公开）；上海AI实验室9/13浦江论坛发布Intern-S2-397B科研向开源基座、vLLM/SGLang Day-0+端砚平台——开源竞争下沉到搜索/科研基础设施层(官方/人民网,高/中高)

英伟达/Palantir限制员工内部用前沿模型

The Information/Reuters2026-09-15置信度高

The Information/路透：英伟达限Claude于低敏感任务、Palantir要求"不可撤销零数据留存"——企业AI治理从对外合规转向对内数据主权(The Information/Reuters,高)

治理密集窗：微软MAI准则+英国王室峰会9/17+欧盟55条首例

Reuters/白金汉宫2026-09-15置信度高

微软MAI《行为准则》9/14公告+6周意见期（修订版年底发布、2027年起指导开发）；英国王室AI峰会9/17邓弗里斯庄园（OpenAI CFO Sarah Friar/黄仁勋/Hassabis出席，Ditchley承办）；欧委会确认按AI法55条审查OpenAI评估代理越界事件——立法/自规/元首三线同周共振(Reuters/白金汉宫,高)

Anthropic连续两季盈利+OpenAI今年不IPO（进展·第3天）

投资者沟通/新浪财经2026-09-15置信度中高进展·第 3 天

投资者披露：Q2营收超115亿首盈利后Q3预计连续盈利（警示重投前沿模型后不保证）；Altman以安全担忧表示OpenAI今年不推进IPO——与Anthropic 10月中旬路演路线分化(投资者沟通/新浪财经,中高)

岚图智算池成交口径修正+沐曦C700接近流片

报道台账2026-09-15

东风采购平台9/15公告：腾讯云（北京）单独中标1619.74万（"8251.30万+三大运营商联合中标"不实，电信/移动湖北系第二三候选人）；沐曦9/7业绩会：C700核心设计与功能验证接近完成、下一步流片，"对标H100"系公司自述无第三方测试(公告/业绩会,高/中)

剔除项存档

报道台账2026-09-15

智谱"A+H首次披露"系6/1已公告（A股150亿募投中80%投通用基座非"60%自进化"；"净募755亿港元"口径拼接）剔除；Epoch AI"86个数据中心13.3GW/累计capex 5020亿/芯片出货30.5M H100e"未获多源核验暂缓；字节OpenViking系8月中旬已开源剔除；IFM K2"完全开源"再报（9/4已报）剔除；优必选乐山中标（9/14已报）无增量不再报

DeepSeek切换落地实测（进展·第4天）

AA实测/多源2026-09-14置信度高进展·第 4 天

AA独立实测V4.1 Flash智能指数39.5反超V4 Pro的36，但单任务平均输出约8.9万token比V4 Pro多62%、部分抵消降价红利；官方"保留原入口"（9/13报）与实测路由并存，迁移以官方文档为准——比价逻辑转向单任务总成本(AA实测/多源,高)

GPT-6 Astra攻克FrontierMath Tier 4

量子位首发/多源印证2026-09-14置信度高

解出该层级全部剩余题97.6%（此前最高约5%），Epoch AI宣布Tier 4饱和；叠加Erdős层2/68上下文，整套FrontierMath被AI完整攻克——科研级基准失去区分度、评测体系加速换代(量子位首发/多源印证,高)

智谱50亿美元融资完成

公告2026-09-14置信度高

9/13公告20亿配售+30亿零息可转债（"小股大债"海外长线抢筹）；ARR16亿美元系8/31半年报已披露口径勿当新增；API毛利率-0.4%→24.6%转正、Token调用较年初+40倍(公告,高)

FOMC前夜：CME加息概率修正口径

CME2026-09-14置信度高

CME 9/14官方口径86.2%（9/13盘后约90%、一周前约59%）；"哈塞特'没有理由加息'原话"未获独立核实剔除；9/16议息+8月零售销售同日落地(CME,高)

AI降速叙事冲击暗盘（进展·第2天）

行情2026-09-14置信度高进展·第 2 天

9/13暗盘SK海力士-4.60%/英特尔-3.81%/美光-3.25%/英伟达-2.16%（与21财经/雪球逐项吻合）；Altman称OpenAI 2026不上市vs Anthropic抢跑史上最大IPO；甲骨文CapEx落点显示算力开支未退潮——情绪与基本面背离(行情,高)

特朗普公开对抗"减速共识"（进展·第2天）

SBN/NewsDrum/Yahoo2026-09-14置信度高进展·第 2 天

9/13爱尔兰表态反对减速AI（"谁赢AI谁赢"）、称风险警告系"负面势力炒作"、只支持护栏；Sanders回击呼吁中美谈判AI暂停条约；Ban ASI Act（9/3预告）攻防白热化(SBN/NewsDrum/Yahoo,高)

Anthropic选定纳斯达克+财务首披露（进展·第2天）

BI/Bloomberg/Reuters2026-09-14置信度高进展·第 2 天

BI/彭博/路透多源：推介材料首次披露Q2营收超115亿美元、首次盈利，Q3"有望"连续盈利；英伟达基石至多100亿仍在洽谈——史上最大IPO进入推介阶段(BI/Bloomberg/Reuters,高)

优必选柳州万台级人形机器人超级工厂投产

金台资讯/科创板日报2026-09-14置信度高

9/12投产：1.4万㎡、每10分钟下线1台WalkerS/Cruzr、年产能超万台（联合西门子数字孪生、零部件国产化率超75%），全球首个万台级基地；另中标乐山1.5081亿(金台资讯/科创板日报,高)

人形机器人H1出货1.91万台+272%

Smart Analytics/科创板日报2026-09-14置信度中高

Smart Analytics Global单一机构口径；高盛预测2035整机均价2.13万美元（-49%）；宇树单价两年降72%至16.64万元；中国厂商占全球出货约97%（产业口径）——Q3出货为下一验证点(Smart Analytics/科创板日报,中高)

Cognition发布SWE-2（Kimi K3后训练）

官方博客2026-09-14置信度高

官方博客证实基于Kimi K3（2.8T）RL后训练，自家FrontierCode 1.1 Main 50.0%、距Fable 5.1不足1分、成本声称低64%；FrontierCode系厂商自建榜单、竞品分数系厂商自测口径打折(官方博客,高)

02官方事件精选OFFICIAL EVENTS20 条

Detecting wildfires with ChatGPT

YouTube·OpenAI2026-09-15置信度高

原文 ↗

Live from Toronto: KlingAI at TIFF The Market

YouTube·Kling AI2026-09-15置信度高

原文 ↗

Share your Managed Deep Agent with your team using Slack

YouTube·LangChain2026-09-15置信度高

原文 ↗

Databricks for Good x Crisis Text Line: Using GenAI to control and optimize costs

YouTube·Databricks2026-09-15置信度高

原文 ↗

UCLA Health Offers Fast, 24/7 Patient Support with Agentforce

YouTube·Salesforce2026-09-15置信度高

原文 ↗

Consistent AI Characters, Voices and Scenes (Advanced Workflow) | Runway

YouTube·Runway2026-09-15置信度高

原文 ↗

Introducing voice, music, image, and video generation in the ElevenLabs MCP

YouTube·ElevenLabs2026-09-15置信度高

原文 ↗

What a NASA astronaut can teach us about discovery and exploration | Christina Koch & James Manyika

YouTube·Google2026-09-15置信度高

原文 ↗

Build a real-time voice AI agent with Google ADK and Gemini Live API

YouTube·Google Cloud2026-09-15置信度高

原文 ↗

What blog posts influenced your thinking the most?

官方·Simon Willison2026-09-15置信度高

原文 ↗

Does everyone have taste? These 5 experts say yes.

YouTube·Canva2026-09-14置信度高

原文 ↗

How Fyxer built an AI executive assistant people trust

YouTube·OpenAI2026-09-14置信度高

原文 ↗

How Developers Secure AI-Generated Code: 5 Security Best Practices

YouTube·IBM Technology2026-09-14置信度高

原文 ↗

Introducing BM25 on AlloyDB & Cloud SQL

YouTube·Google Cloud2026-09-14置信度高

原文 ↗

03榜单异动RANKING MOVES8 条

multica-ai/andrej-karpathy-skills 星标 +76,065

GitHub2026-09-01 → 2026-09-10

原文 ↗

Shubhamsaboo/awesome-llm-apps 星标 +49,199

GitHub2026-09-01 → 2026-09-10

原文 ↗

farion1231/cc-switch 星标 +47,426

GitHub2026-09-01 → 2026-09-10

原文 ↗

nextlevelbuilder/ui-ux-pro-max-skill 星标 +45,383

GitHub2026-09-01 → 2026-09-10

原文 ↗

Graphify-Labs/graphify 星标 +41,752

GitHub2026-09-01 → 2026-09-10

原文 ↗

生态脉搏站点数据自更新

2026-09-01 → 2026-09-10 星标涨幅：obra/superpowers +101,959★ · mattpocock/skills +92,450★ · multica-ai/andrej-karpathy-skills +76,065★ · anthropics/skills +63,075★ · Shubhamsaboo/awesome-llm-apps +49,199★ · farion1231/cc-switch +47,426★ · nextlevelbuilder/ui-ux-pro-max-skill +45,383★ · Graphify-Labs/graphify +41,752★

※风险分级灰犀牛 / 黑天鹅

灰犀牛 · 可预见

灰犀牛Ⅰ FOMC 鹰派落地引发 AI 高估值二次杀跌｜概率 90-92%（CME 定价）｜冲击面：AI 高估值 / 算力链｜观察哨：9/16 凌晨决议 + 点阵图（深度版风险分级）
灰犀牛Ⅱ 「注意义务」法案进入表决流程：头部公司合规成本激增、模型发布节奏放缓｜概率：低-中（原文口径：仍为草案、成法概率低）｜冲击面：前沿模型供给 / 合规成本｜观察哨：参议院正式文本 / 委员会标记（深度版风险分级）
灰犀牛Ⅲ capex 与营收错配：算力资本开支高企下若收入增速不匹配，2027 年折旧墙提前计价｜概率：高概率·可预见（深度版分类）｜冲击面：算力链毛利｜观察哨：SaaS / 云三季报毛利率（深度版风险分级）

黑天鹅 · 低概率大冲击

黑天鹅Ⅰ 欧盟 AI 法第 55 条审查升级为对 OpenAI 智能体服务的行政限制令｜概率：低概率·高冲击（深度版分类）｜冲击面：Agent 服务合规｜观察哨：欧委会后续处置（深度版风险分级）
黑天鹅Ⅱ 头部实验室因「自证防伤害」要求集中延迟发布，引发 IPO 估值重定价｜概率：低概率·高冲击（深度版分类）｜冲击面：一级估值 / 融资环境｜观察哨：10 月中旬 Anthropic 路演认购倍数（深度版风险分级）

✦行动建议3

1. Agent 厂商：把交付从单点做成整包（同业复制期）

锁定银行标书 Agent / 券商投研订阅等已爆发场景，改造成整包交付（数据合规 + 审计日志 + SLA 打包报价），对标浙商 7436 万 / 月、国投 730 天订阅制采购结构。进阶：先用央企「小额高频」标攒案例，再跨同业复制（银行方案平移券商 / 保险，复用率 >70%）。打包：接入国产模型（GLM / DeepSeek）降本，成本差即报价空间。

时间窗时间窗：1-2 个季度内先入者锁定行业模板

2. 合规工具包：把「治理密集窗」变成产品需求

做 AI 合规最小产品——模型调用日志 + 生成内容审计 + 身份认证三件套，先服务被采购倒逼的金融 Agent 厂商；对标欧盟 AI 法 55 条 / 美「注意义务」草案做 checklist 化评估工具，卖「合规订阅」而非项目制。与①捆绑：Agent 整包自带合规模块，一鱼两吃。

时间窗时间窗：执法落地前 2-4 个季度（9/16 吹风 + 立法草案为免费需求教育）

3. 独立开发者：垂直开源 Agent 卡位（需快速验证）

基于 Iris / 书生-S2 做行业特化封装（如面向中小商家的内容搜索 Agent），两周内 MVP 上线收费验证；通过 → 行业模板库，不通过 → 转为①的引流钩子。

时间窗时间窗：约 2 个季度（只投入可承受的验证成本）

⏱未来 3 个月时间线概率

时间窗	事件	概率	依据
9/15-16	FOMC 议息（加息 90-92%）+ 8 月零售销售同日落地	90-92% 加息	↓ 风险资产；软件/安全相对抗跌（CME + 深度版时间线）
9/16	Ban ASI Act 吹风会	高	↑ 监管叙事强化，攻击性条款难成法（深度版时间线）
9/17	英国王室 AI 峰会（邓弗里斯庄园，已官宣）	高	↑ 治理密集窗顶点，中期利多合规板块（深度版时间线）
9/24	长江证券智能体云服务投标截止	高	验证：券商采购落地验证点（公告日程）
9/29	OpenAI DevDay + 上海公安智能体平台开标	高	↑ Agent 平台政务定价锚（已公告 + 临近）
9/30	加州 30 项法案签署窗口 + ARC-AGI-3 截榜	高	规则与评测双拐点（法定窗口）
10/1	OpenAI vs Apple 听证会	高	↑ Agent 入口之争公开化（听证会日程）
10 月中旬	Anthropic IPO 路演（约 2 万亿估值）	中高	↑ 一级估值锚上移，带动 AI 股情绪（深度版时间线）
10-11 月	美「注意义务」法案委员会审议	中	↑ 监管溢价，模型发布减速（草案流程）
11 月	人形机器人 Q3 出货验证 + 智谱资金到账放量	中高	验证：产业兑现验证点（Q3 数据 + 公告节奏）
12 月	费半调整后资金回流硬件链（技术性反弹窗口）	中	硬件链逢低配置窗口（深度版时间线）

素材来源：报道台账（滚动查重）：2026-09-01 ~ 2026-09-15 共 208 条主题记录 · AI 官方视频监控台账：15 天 · 429 条事件（精选 20 条带链接） · 站点榜单快照：2026-09-01 → 2026-09-10 星标涨幅 Top8 · 风险/行动/时间线：承接 2026-09-15 日报
SkillHub.news — 全球 Skill 榜单

MONTH 2026-09 · 52 STORIES · SKILLHUB MONTHLY

SkillHub.news Monthly AI

2026-09 · Monthly · sources 4

Highlights52 stories · approx.21min read

MONTHLY THREADS24 stories

SOX -5.86%: First Pricing Day for Slowdown Calls

OFFICIAL EVENTS20 stories

Detecting wildfires with ChatGPT

RANKING MOVES8 stories

obra/superpowers stars +101,959

01MONTHLY THREADS24 stories

SOX -5.86%: First Pricing Day for Slowdown Calls

多源2026-09-15Confidence High

9/14 SOX -5.86% (largest single-day drop since 7/1), NVIDIA -3.36%/Micron -5.25%/Intel -5.59%/Corning -13.7%; Nasdaq only -0.56%, CrowdStrike +13.85% hit a new high, software stocks rose 4-7%—same-day hardware→software/security rotation completed, first market confirmation of AI narrative switch (multi-source, high)

U.S. Senate 'duty of care' bill draft vs Trump veto

Reuters2026-09-15Confidence High

Thune/Cruz/Klobuchar discuss: require AI companies to self-certify harm prevention + authorize Commerce Secretary to send reviewers to test products; still a draft, low chance of becoming law; Trump 9/14 hinted he won't sign—federal 'legislative vs executive' split takes shape, resonating same week with 9/16 Ban ASI Act briefing (Reuters, high)

Chinese open-source models climb on all three leaderboards (GLM-5.3 + DeepSeek V4.1-Flash)

SuperCLUE/AA官网2026-09-15Confidence High

SuperCLUE September edition (official site 9/11 batch): GLM-5.3 71.29 (+8.02) jumped from 5th to 3rd, V4.1-Flash 71.81 debuts as No. 2 domestic; AA Intelligence Index GLM-5.3 44.9 No. 1 open-source (K3 43.8/Flash 39.5); LMArena company leaderboard GLM-5.3 5th—three metrics corroborate each other, post-training + architecture refresh, not a pretraining breakthrough (SuperCLUE/AA official sites, high)

OpenRouter new week 127 trillion: China surpasses US for 20 consecutive weeks

OpenRouter/每经2026-09-15Confidence High

9/7-9/13 global 126.8-127 trillion tokens (+10% period-over-period), 5.55 billion API calls; China 61.17 trillion vs US 21.76 trillion; Xiaomi MiMo-V2.5 +230% entered top five; V4.1 Flash ranked 6th 3 days after launch; official and aggregator leaderboards have different statistical windows, so rankings are inconsistent; citations must note the window (OpenRouter/National Business Daily, high)

Macro tightening on three fronts: 10Y first breaks 5% + pipeline attacked + FOMC 90-92% (Progress · Day 1)

新华社/CME2026-09-15Confidence HighProgress · Day 1

10Y hit 5.017% intraday (first breach since Oct 2023); Saudi oil pipeline attacked, damaged, and halted (not a voluntary shutdown); Brent closed 105.68/WTI 101.39 (intraday 109.8/104.95); CME rate-hike odds rose to 90-92% (86.2% on morning 9/14); tonight FOMC + August retail sales both land same day (Xinhua/CME, High)

OpenAI pretraining safety protocol lands (Progress · Day 3)

OpenAI官方2026-09-15Confidence HighProgress · Day 3

Official blog + Altman on X: frontier RL runs must establish safety cases before launch (safety case threshold) — the verbal consensus of "Dario long post → Altman endorsement" was codified into protocol text for the first time; leading labs' version schedules add an extra safety gate (OpenAI official, High)

Jiuwanli Future Technology raises over $100 million seed + angel round

9/14多方报道2026-09-15Confidence Mid

Founded by Chen Peng, former Horizon Robotics chip president; edge/on-device high-compute Agentic inference chips; with participation from Sequoia China, Huaye Tiancheng, BlueRun Ventures, Lenovo Capital, CAS Star, and Wuyuefeng; Gaohu as exclusive FA — primary-market money continues to flow directly into chips/compute infrastructure (multiple reports on 9/14, medium-high)

Brokerage AI agent procurement surges + 730-day subscription model

报道台账2026-09-15

Changjiang Securities financial dialogue agent cloud service tender (9/8 announcement, 9/14 registration deadline, 9/24 bid deadline); SDIC Securities AI smart workbench authorization service agreement 730 days (9/3 announcement); in August, 9 brokerages had 10 requirements spanning compliance/trading/investment research; banking diffusion: Chongqing Rural Commercial Bank's two lots exceed 10 million—AI procurement enters peer replication phase (announcement/Smart Hyperparameters, high/medium-high)

Xiaohongshu open-sources Search Agent Iris+ and Shusheng-S2 research foundation model

报道台账2026-09-15

Iris-mini 35B/Iris-pro 397B (based on Qwen; strongest open-source in its class on BrowseComp at 82.2/88.6; weights + eval code released); Shanghai AI Lab launched Intern-S2-397B, a research-oriented open-source foundation model, at the 9/13 Pujiang Forum, with vLLM/SGLang Day-0+ and DuanYan platform — open-source competition shifts down to the search/research infrastructure layer (official/People.cn, high/medium-high)

NVIDIA/Palantir restrict employees' internal use of frontier models

The Information/Reuters2026-09-15Confidence High

The Information/Reuters: NVIDIA restricts Claude to low-sensitivity tasks; Palantir demands 'irrevocable zero data retention' — enterprise AI governance shifts from external compliance to internal data sovereignty (The Information/Reuters, High)

Dense governance window: Microsoft MAI guidelines + UK royal summit 9/17 + first EU Article 55 case

Reuters/白金汉宫2026-09-15Confidence High

Microsoft MAI Code of Conduct announced 9/14 + 6-week comment period (revised version to be released by year-end, guiding development from 2027); UK royal AI summit 9/17 at Dumfries House (OpenAI CFO Sarah Friar/Jensen Huang/Hassabis attending, hosted by Ditchley); European Commission confirms review of OpenAI evaluation agent overreach incident under Article 55 of the AI Act — legislative/self-regulatory/head-of-state tracks resonate in the same week (Reuters/Buckingham Palace, High)

Anthropic profitable for two consecutive quarters + OpenAI not to IPO this year (Progress · Day 3)

投资者沟通/新浪财经2026-09-15Confidence MidProgress · Day 3

Investor disclosure: Q2 revenue over 115亿 and first profit; Q3 expected to remain profitable (warns no guarantee after heavy reinvestment in frontier models); Altman, citing safety concerns, says OpenAI will not pursue IPO this year—diverging from Anthropic's mid-October roadshow path (investor communications/Sina Finance, medium-high)

Voyah intelligent computing pool transaction basis corrected + MetaX C700 nears tape-out

报道台账2026-09-15

Dongfeng procurement platform 9/15 announcement: Tencent Cloud (Beijing) won bid alone for 1619.74万 ("8251.30万 + three major operators joint win" false; Telecom/Mobile Hubei are second and third candidates); MetaX 9/7 earnings call: C700 core design and functional verification nearly complete, next step tape-out; "benchmarks H100" is company claim with no third-party testing (announcement/earnings call, high/medium)

Excluded items archive

报道台账2026-09-15

Zhipu's "A+H first disclosure" was already announced on 6/1 (of A-share 150亿 fundraising, 80% goes to general foundation, not "60% self-evolution"; "net raise 755亿港元" is spliced basis) removed; Epoch AI's "86 data centers 13.3GW/cumulative capex 5020亿/chip shipments 30.5M H100e" not multi-source verified, deferred; ByteDance OpenViking open-sourced in mid-August, removed; IFM K2 "fully open source" repeated (already reported 9/4), removed; UBTECH Leshan win (already reported 9/14) no increment, no longer reported

DeepSeek switchover rollout test (Day 4 progress)

AA实测/多源2026-09-14Confidence HighProgress · Day 4

AA independent tests show V4.1 Flash's intelligence index at 39.5, overtaking V4 Pro's 36, but average output per task is about 89,000 tokens, 62% more than V4 Pro, partly offsetting the price-cut benefit; the official 'keep original entry point' (9/13 report) coexists with tested routing, and migration is subject to official documentation—price-comparison logic shifts to total cost per task (AA tests/multiple sources, high)

GPT-6 Astra cracks FrontierMath Tier 4

量子位首发/多源印证2026-09-14Confidence High

Solved 97.6% of all remaining problems at that tier (previous high about 5%), Epoch AI declares Tier 4 saturated; with the Erdős layer 2/68 context added, the entire FrontierMath suite has been fully cracked by AI—research-grade benchmarks lose discrimination, and the evaluation system accelerates generational turnover (QbitAI first report/multiple sources corroborate, high)

Zhipu completes $5 billion financing

公告2026-09-14Confidence High

9/13 announcement: 2 billion placement + 3 billion zero-coupon convertible bonds ('small equity, large debt'; overseas long-term investors scramble for shares); ARR $1.6 billion is the already-disclosed 8/31 half-year report basis—do not treat as new; API gross margin -0.4%→24.6% turned positive, Token calls +40x vs. start of year (announcement, high)

FOMC eve: CME rate-hike probability revised basis

CME2026-09-14Confidence High

CME 9/14 official reading 86.2% (about 90% after the 9/13 close, about 59% a week earlier); Hassett's exact quote 'no reason to hike rates' was not independently verified and removed; 9/16 rate meeting + August retail sales land same day (CME, high)

AI slowdown narrative hits dark pool (Progress · Day 2)

行情2026-09-14Confidence HighProgress · Day 2

9/13 dark pool: SK Hynix -4.60%/Intel -3.81%/Micron -3.25%/NVIDIA -2.16% (matches 21财经/Xueqiu item by item); Altman says OpenAI won't go public in 2026 vs Anthropic races ahead on largest-ever IPO; Oracle CapEx allocation shows compute spending has not ebbed—sentiment diverges from fundamentals (Markets, High)

Trump openly confronts 'deceleration consensus' (Progress · Day 2)

SBN/NewsDrum/Yahoo2026-09-14Confidence HighProgress · Day 2

9/13 Ireland voices opposition to AI deceleration ('Whoever wins AI wins'), calls risk warnings 'hype by negative forces', supports only guardrails; Sanders hits back, calls for U.S.-China negotiations on an AI pause treaty; battle over Ban ASI Act (previewed 9/3) intensifies (SBN/NewsDrum/Yahoo, High)

Anthropic selects Nasdaq + first financial disclosure (Progress · Day 2)

BI/Bloomberg/Reuters2026-09-14Confidence HighProgress · Day 2

BI/Bloomberg/Reuters multiple sources: pitch materials disclose for the first time Q2 revenue over $11.5 billion, first profit; Q3 "expected" to be profitable again; NVIDIA cornerstone investment of up to $10 billion still in talks—largest-ever IPO enters pitch stage (BI/Bloomberg/Reuters, high)

UBTECH's Liuzhou 10,000-unit-class humanoid robot super factory begins production

金台资讯/科创板日报2026-09-14Confidence High

Production starts 9/12: 14,000 m²; one WalkerS/Cruzr rolls off the line every 10 minutes; annual capacity over 10,000 units (with Siemens digital twin; component localization rate over 75%); world's first 10,000-unit-class base; also wins Leshan bid of 150.81 million (Jintai Information/STAR Market Daily, high)

Humanoid robot H1 shipments 19,100 units, +272%

Smart Analytics/科创板日报2026-09-14Confidence Mid

Smart Analytics Global single-institution basis; Goldman Sachs forecasts 2035 average whole-robot price at $21,300 (-49%); Unitree unit price down 72% over two years to RMB 166,400; Chinese vendors account for about 97% of global shipments (industry basis)—Q3 shipments are the next verification point (Smart Analytics/科创板日报, medium-high)

Cognition releases SWE-2 (post-trained on Kimi K3)

官方博客2026-09-14Confidence High

Official blog confirms RL post-training based on Kimi K3 (2.8T); its own FrontierCode 1.1 Main 50.0%, less than 1 point behind Fable 5.1, claimed cost 64% lower; FrontierCode is a vendor-built leaderboard, and competitor scores are discounted based on vendor self-testing (official blog, high)

02OFFICIAL EVENTS20 stories

Detecting wildfires with ChatGPT

YouTube·OpenAI2026-09-15Confidence High

Original ↗

What is an agent harness?

YouTube·Google Cloud2026-09-15Confidence High

Original ↗

Live from Toronto: KlingAI at TIFF The Market

YouTube·Kling AI2026-09-15Confidence High

Original ↗

Share your Managed Deep Agent with your team using Slack

YouTube·LangChain2026-09-15Confidence High

Original ↗

Databricks for Good x Crisis Text Line: Using GenAI to control and optimize costs

YouTube·Databricks2026-09-15Confidence High

Original ↗

UCLA Health Offers Fast, 24/7 Patient Support with Agentforce

YouTube·Salesforce2026-09-15Confidence High

Original ↗

Consistent AI Characters, Voices and Scenes (Advanced Workflow) | Runway

YouTube·Runway2026-09-15Confidence High

Original ↗

Introducing voice, music, image, and video generation in the ElevenLabs MCP

YouTube·ElevenLabs2026-09-15Confidence High

Original ↗

What a NASA astronaut can teach us about discovery and exploration | Christina Koch & James Manyika

YouTube·Google2026-09-15Confidence High

Original ↗

Build a real-time voice AI agent with Google ADK and Gemini Live API

YouTube·Google Cloud2026-09-15Confidence High

Original ↗

What blog posts influenced your thinking the most?

官方·Simon Willison2026-09-15Confidence High

Original ↗

Customer Stories - Juicebox

YouTube·Fireworks AI2026-09-15Confidence High

Original ↗

The contagion of fear

官方·Simon Willison2026-09-15Confidence High

Original ↗

Does everyone have taste? These 5 experts say yes.

YouTube·Canva2026-09-14Confidence High

Original ↗

shot-scraper 1.12

官方·Simon Willison2026-09-14Confidence High

Original ↗

commit-rewriter 0.1

官方·Simon Willison2026-09-14Confidence High

Original ↗

梦主

B站·可灵AI2026-09-14Confidence High

Original ↗

How Fyxer built an AI executive assistant people trust

YouTube·OpenAI2026-09-14Confidence High

Original ↗

How Developers Secure AI-Generated Code: 5 Security Best Practices

YouTube·IBM Technology2026-09-14Confidence High

Original ↗

Introducing BM25 on AlloyDB & Cloud SQL

YouTube·Google Cloud2026-09-14Confidence High

Original ↗

03RANKING MOVES8 stories

obra/superpowers stars +101,959

GitHub2026-09-01 → 2026-09-10

Original ↗

mattpocock/skills stars +92,450

GitHub2026-09-01 → 2026-09-10

Original ↗

multica-ai/andrej-karpathy-skills stars +76,065

GitHub2026-09-01 → 2026-09-10

Original ↗

anthropics/skills stars +63,075

GitHub2026-09-01 → 2026-09-10

Original ↗

Shubhamsaboo/awesome-llm-apps stars +49,199

GitHub2026-09-01 → 2026-09-10

Original ↗

farion1231/cc-switch stars +47,426

GitHub2026-09-01 → 2026-09-10

Original ↗

nextlevelbuilder/ui-ux-pro-max-skill stars +45,383

GitHub2026-09-01 → 2026-09-10

Original ↗

Graphify-Labs/graphify stars +41,752

GitHub2026-09-01 → 2026-09-10

Original ↗

Ecosystem Pulseself-updating

2026-09-01 → 2026-09-10 Star growth：obra/superpowers +101,959★ · mattpocock/skills +92,450★ · multica-ai/andrej-karpathy-skills +76,065★ · anthropics/skills +63,075★ · Shubhamsaboo/awesome-llm-apps +49,199★ · farion1231/cc-switch +47,426★ · nextlevelbuilder/ui-ux-pro-max-skill +45,383★ · Graphify-Labs/graphify +41,752★

※Risk Levelsgray rhino / black swan

Gray rhino · foreseeable

Gray Rhino I: FOMC hawkish outcome triggers second selloff in high-valuation AI | Probability 90-92% (CME pricing) | Impact: high-valuation AI / compute chain | Watch: 9/16 early-morning decision + dot plot (deep-dive risk grading)
Gray Rhino II: 'Duty of Care' bill enters voting process: compliance costs for leading companies surge, model release pace slows | Probability: low-medium (source wording: still a draft, low chance of becoming law) | Impact: frontier model supply / compliance costs | Watch: Senate final text / committee markup (deep-dive risk grading)
Gray Rhino III: capex-revenue mismatch: with compute capex high, if revenue growth doesn't match, 2027 depreciation wall priced in early | Probability: high-probability·foreseeable (deep-dive classification) | Impact: compute chain gross margins | Watch: SaaS / cloud Q3 report gross margins (deep-dive risk grading)

Black swan · low-probability

Black Swan I: EU AI Act Article 55 review escalates into an administrative restriction order on OpenAI agent services | Probability: low-probability·high-impact (deep-dive classification) | Impact: Agent service compliance | Watch: European Commission follow-up action (deep-dive risk grading)
Black Swan II: Leading labs delay releases en masse due to 'self-certifying harm prevention' requirements, triggering IPO valuation repricing | Probability: low-probability·high-impact (deep-dive classification) | Impact: primary-market valuation / financing environment | Watch: mid-October Anthropic roadshow subscription multiple (deep-dive risk grading)

✦Action Items3

1. Agent vendors: move delivery from point solutions to full packages (peer-replication phase).

Target already-hot scenarios such as bank bid-document Agent / brokerage investment-research subscriptions, and turn them into full-package delivery (data compliance + audit logs + SLA bundled pricing), benchmarking 浙商 7436 万 / 月 and 国投 730 天 subscription procurement structures. Next: use central-SOE “small-amount, high-frequency” tenders to build cases, then replicate across peers (port bank solutions to brokerages / insurance; reuse rate >70%). Bundle: integrate domestic models (GLM / DeepSeek) to cut costs, with the cost gap as pricing headroom.

window Time window: first movers lock in industry templates within 1-2 quarters.

2. Compliance toolkit: turn the “governance-intensive window” into product demand.

Build a minimal AI compliance product—model invocation logs + generated-content audit + identity authentication trio, initially serving financial Agent vendors forced by procurement; benchmark against EU AI Act Article 55 / US “duty of care” draft to make a checklist-style assessment tool, selling “compliance subscriptions” rather than project-based work. Bundle with ①: Agent full packages include built-in compliance modules, a two-for-one play.

window Time window: 2-4 quarters before enforcement lands (9/16 signaling + legislative drafts as free demand education).

3. Indie developers: stake a position in vertical open-source Agent (needs rapid validation).

Based on Iris / 书生-S2, build industry-specific wrappers (e.g., a content search Agent for small and medium merchants); launch MVP within two weeks for paid validation; if it passes → industry template library, if not → turn it into the lead-gen hook in ①.

window Time window: about 2 quarters (invest only affordable validation costs).

⏱Next 3 MonthsProb.

Window	Event	Prob.	Basis
9/15-16	FOMC rate decision (rate hike 90-92%) + August retail sales land same day.	90-92% rate hike.	↓ risk assets; software/security relatively resilient (CME + deep-dive timeline).
9/16	Ban ASI Act briefing.	High.	↑ regulatory narrative strengthens; aggressive provisions unlikely to become law (deep-dive timeline).
9/17	British Royal AI Summit (Dumfries House, officially announced)	High	↑ Peak of governance-intensive window; medium-term bullish for compliance sector (in-depth timeline)
9/24	Changjiang Securities AI agent cloud service bidding deadline	High	Verification: brokerage procurement implementation checkpoint (announcement schedule)
9/29	OpenAI DevDay + Shanghai Public Security agent platform bid opening	High	↑ Agent platform government-affairs pricing anchor (announced + approaching)
9/30	California 30-bill signing window + ARC-AGI-3 leaderboard cutoff	High	Rules and evaluation at dual inflection points (statutory window)
10/1	OpenAI vs Apple hearing	High	↑ Agent entry-point battle goes public (hearing schedule)
10 月中旬	Anthropic IPO roadshow (about 2 trillion valuation)	Medium-high	↑ primary-market valuation anchor moves up, lifting AI stock sentiment (in-depth timeline)
10-11 月	U.S. 'duty of care' bill committee review	Medium	↑ Regulatory premium, model release slowdown (draft process)
11 月	Humanoid robot Q3 shipment validation + Zhipu funding received, volume ramp	Medium-High	Validation: industry delivery checkpoints (Q3 data + announcement cadence)
12 月	Capital returns to hardware chain after SOX correction (technical rebound window)	Medium	Hardware chain dip-buying allocation window (deep-dive timeline)

Sources：Coverage ledger (rolling deduplication)：2026-09-01 ~ 2026-09-15, 208 topic records total · AI Official Video Monitoring Log：15 days · 429 events (20 selected with links) · Site leaderboard snapshot：2026-09-01 → 2026-09-10 Star Growth Top8 · Risk/Action/Timeline：Follows the 2026-09-15 daily report
SkillHub.news — Global Skill rankings