报告形态 · 五维分类 + 风险 + 行动 + 时间线 ← 回站点
WEEK 2026-09-13 · 46 STORIES · SKILLHUB WEEKLY
SkillHub.news 每周AI
9 月 7 日 ~ 2026 年 9 月 13 日 · 周报 · 素材 4
本期看点46 条 · 约18分钟
01
本周主题线WEEKLY THREADS24 条
DeepSeek撤回V4 Pro下线(新拐点,反转9/12切换)
02
官方事件精选OFFICIAL EVENTS14 条
Quoting Paul Ford
03
榜单异动RANKING MOVES8 条
obra/superpowers 星标 +79,301
01本周主题线WEEKLY THREADS24 条

DeepSeek撤回V4 Pro下线(新拐点,反转9/12切换)

报道台账2026-09-13
官方9/11晚邮件+9/12官宣:9/14后保留原API入口、计费不变、不强制路由Flash,企业可自选版本——开发者舆情三天逼回商业决策,API生命周期话语权首次转向用户(官方邮件/AI Model Watch/tokencost三方一致,高)

Dario《We Must Pace the Frontier》+Altman附议

Reuters等2026-09-13置信度 高
9/12-13凌晨发文:三步计划+Anthropic单方面承诺第三方评估永久员工级访问,Altman数小时后"we will do the same"、Musk亦表态;时点微妙——Anthropic冲刺约2万亿估值IPO,"放缓"与最大IPO同框(Reuters等,高)

25位菲尔兹奖得主联署抨击难题基准

多源2026-09-13置信度 中高
陶哲轩/舒尔茨/德利涅等25位9/11联署《A Severe Misalignment of AI in Mathematics》,约2500人跟进签名;导火索与9/9 Navier-Stokes争议同源——学界与实验室评测合法性之争公开化(多源,中高)

甲骨文CapEx落点:戴尔HPE双新高(进展·第2天)

业绩会/行情2026-09-13置信度 高进展·第 2 天
CFO业绩会重申FY2027资本开支900-950亿美元并罕见点名戴尔/HPE承接AI机架/液冷/网络设备;9/11戴尔+11.98%市值破3600亿创新高、HPE+约10.5-11%、超微订单簿600亿——AI基建首次有"收货人名单"(业绩会/行情,高)

电信研究院Token报告:2026年10亿亿

报道台账2026-09-13
《智能体时代AI基础设施发展研究报告(2026)》:2026中国Token年消耗10亿亿(10^18)、2030超3500亿亿;2029推理算力占比80%;全球活跃智能体2026约8000万个、2030预计22.16亿个(央视转引,中·权威单报告)

央企算力集采单日三单近9亿

财富号转述待公告核2026-09-13置信度 中
9/12:中国长城中标国网第三期服务器超3亿+软通计算机中标移动部件焕新4.18亿+平知信息预中标云南智算约1.37亿(财富号转述待公告核,中);同期京津冀"太空算力长廊"倡议+北京超1亿生物医药算力专项(一年赋能100项目)

Real-SWE私有代码库基准首榜

多源2026-09-13置信度 中高
考题全部来自授权私有生产库(计费/税务/迁移,参考解中位改11文件);Fable5.1+Claude Code 38.8%/GPT-6 Astra+Codex CLI 33.8%/Gemini3.8 31.2%/GLM5.3 28.8%——分数绑定模型+工具链,harness成排名前提(多源,中高)

SREGym:SRE智能体差两成闭环

单源2026-09-13置信度 中
GPT-5.6 Sol(max)+Codex 81.0%端到端解决率居首(诊断95.2%/缓解85.7%)、Claude Opus 5+Claude Code 76.2%;最佳配置约两成生产故障无法闭环、配置间Token成本差近2倍(单源,中)

智能体治理空白带(商机)

三源2026-09-13置信度 高
OpenAI智能体滥用RubyDoc致RubyGems停注册4天(HN 9/12)+48%已部署智能体缺有效安全控制+外滩大会报告判治理为"必要条件"——身份认证/权限管理/行为审计SaaS中美共振空白(三源,高)

资本下沉"卖水人"三连投

报道台账2026-09-13置信度 中
Enfabrica 1.25亿美元(英伟达战略参投互连芯片)/烨知心成立一年累计近10亿元(端侧NPU,深创投等联投,团队出自苹果ANE)/悦点科技FDE数千万Pre-A(自报客户近100家盈亏平衡,公司口径);Mecka AI 5亿美元估值单源暂缓(科创板日报,高/中)

九部门智驾"十五五"规划

工信部等2026-09-13置信度 高
9/9印发9/11公布:2030自动驾驶规模应用+高速城快路高度自动驾驶;明确"搭载自动驾驶系统车辆安全性能大幅超过人类驾驶员"硬指标;加快自动驾驶/自动泊车安全强制国标——智驾安全首次写成产业规划治理类硬指标(工信部等,高)

前瞻日历更新(9月密集窗)

多源2026-09-13置信度 高
9/14 Siri AI随iOS 27 beta上线+Claude Code周限额下调17%生效;9/29 OpenAI DevDay+上海公安1212.5万开标;9/30加州法案窗口+ARC-AGI-3截榜;10/1 OpenAI vs Apple听证会;10月中旬Anthropic IPO路演(多源,高)

剔除项存档

报道台账2026-09-13置信度 中
Qwen3-Next-80B开源系2025年9月旧闻回炒("阿里美股+8%"不实,实+0.68%)剔除;智谱GLM"9/12全球开放"日期口径错(GLM-5.3系8/14发布、8/28开源);英伟达100亿投Anthropic与9/12"拟100亿基石"重复;智谱ARR16亿系9/5已报口径、MiniMax ARR8亿无来源;美股终结四连跌9/12已报;grok-4.6 Fullstack登顶仅镜像站单源不采信;Mecka AI估值单源暂缓;版权中心肖像声纹细则/加拿大合规指南仅播客转述暂缓周一核

PaperCut漏洞武器化AI代理(395组织/48国)

GreyNoise/Blackpoint2026-09-12置信度 高
GreyNoise+Blackpoint双源:利用CVE-2026-81578/82078让Codex(harness)+DeepSeek模型代理自动化开发执行攻击,数小时攻破48国395家组织、最快7分钟域管(440实例中12个全域管,引用带口径)——代理武器化首例规模实战(GreyNoise/Blackpoint,高)

五角大楼拟贷50亿美元给Fluidstack

WSJ/Reuters2026-09-12置信度 高
WSJ 9/10首发+路透证实:经战略资本办公室(OSC)向云初创Fluidstack提供约50亿贷款用于数据中心零部件供应链(非新建),谈判未定案——AI基建首次出现国家信用直接兜底融资(WSJ/Reuters,高)

OpenAI暂停Pro 20X新订阅

OpenAI官宣/PCMag2026-09-12置信度 高
产品负责人Sottiaux 9/10 X官宣:GPT-6 Astra需求"前所未见",$200/月20X档停新订与升级、存量不受影响——旗舰订阅被自家需求挤爆,与Token账本(日推理费>600美元)互证(OpenAI官宣/PCMag,高)

DeepSeek 9/14旗舰退役式切换+HBM需求1/4(进展·第2天)

DeepSeek官方+行情2026-09-12置信度 高进展·第 2 天
官方确认9/14起V4 Pro全量路由至V4.1 Flash按Flash计价(模型卡Terminal-Bench 4.0=31.2);新架构KV cache对HBM需求仅旧版1/4、SSD 1/8,SK海力士美股-5.2%/美光-4.9%——算法效率首次反向冲击存储涨价叙事;AA 39.55/LiveBench 77.3未独立核到不采信(DeepSeek官方+行情,高)

月之暗面ARR破10亿冲20亿(进展·第1天)

Bloomberg2026-09-12置信度 高进展·第 1 天
彭博9/11:Kimi K3带动8月ARR达10亿美元(6月3亿、5周+2.3倍),年底冲20亿、按500亿投前估值推进融资——为A+H上市线补收入端验证(Bloomberg,高)

工信部"AI+软件"专项行动方案

工信部/新华社2026-09-12置信度 高
信发〔2026〕209号(9/2印发、9/11公开发布):2028年覆盖2万家规上软件企业+100项技改+100个智能体标杆+5个开源项目、配套算力券;首次系统部署AI生成代码安全审查与智能体身份标识(工信部/新华社,高)

Anthropic IPO前移10月+OpenAI秘密递表

Reuters/FT转述2026-09-12置信度 中高
路透9/12证实Anthropic 10月中旬路演、中期选举前完成、募资至多1000亿/估值约2万亿、英伟达拟100亿基石;多源证实OpenAI已于6/8秘密递交S-1——安全叙事与资本叙事同框(Reuters/FT转述,中高)

8月CPI落地:软件分项+25.4%进通胀口径(进展·第2天)

BLS/CME2026-09-12置信度 高进展·第 2 天
整体3.4%符合预期、核心环比+0.3%超预期(4月来最大),9/16加息25bp概率尾盘约87%-90%(CME不同时点读数);CPI分项"消费软件及配件"同比+25.4%创有记录新高;美股利空落地收涨终结四连跌(BLS/CME,高)

加州SB 867禁AI陪伴玩具+ASI法案罚则

加州官方/fact sheet2026-09-12置信度 高
纽森签署SB 867:全美首例禁16岁以下儿童玩具搭载AI陪伴聊天功能(5年、无披露豁免路径);Sanders《禁止ASI法案》框架:企业解散式"死刑"+个人至多20年(9/3提案、9/16吹风,细节以正式文本为准)(加州官方/fact sheet,高)

蚂蚁定调智能体商业时刻+涌现奖

第一财经2026-09-12置信度 高
韩歆毅9/11外滩大会"智能体商业ChatGPT时刻"、点名"商家没有好的智能体服务";支付宝每年1000万元"智能体涌现奖"(单项最高100万)——AEO(Agent可调用改造)成空白带(第一财经,高)

评测范式转向harness:HarnessDev+方升-Code 2.0

报道台账2026-09-12
字节Seed等发HarnessDev:同权重换harness分差近15点(GPT-5 Terminus 2 35.2% vs Codex CLI 49.6%);信通院方升-Code 2.0(5000+真实项目数据、新增工程质量/稳定性维度)(encorp转述/腾讯新闻,中/高)
02官方事件精选OFFICIAL EVENTS14 条

Quoting Paul Ford

官方·Simon Willison2026-09-13置信度 高
原文 ↗

Introducing Music v2.5

YouTube·ElevenLabs2026-09-12置信度 高
原文 ↗

Quoting huggingface.co/security.txt

官方·Simon Willison2026-09-12置信度 高
原文 ↗

Soft-deprecating re.match()

官方·Simon Willison2026-09-12置信度 高
原文 ↗

GPT-Live-1 is now in the API

YouTube·OpenAI2026-09-12置信度 高
原文 ↗

Quoting Boris Cherny

官方·Simon Willison2026-09-12置信度 高
原文 ↗
03榜单异动RANKING MOVES8 条

obra/superpowers 星标 +79,301

GitHub2026-09-03 → 2026-09-10
原文 ↗

mattpocock/skills 星标 +71,905

GitHub2026-09-03 → 2026-09-10
原文 ↗

anthropics/skills 星标 +49,058

GitHub2026-09-03 → 2026-09-10
原文 ↗

farion1231/cc-switch 星标 +36,887

GitHub2026-09-03 → 2026-09-10
原文 ↗
生态脉搏站点数据自更新
2026-09-03 → 2026-09-10 星标涨幅:obra/superpowers +79,301★ · mattpocock/skills +71,905★ · multica-ai/andrej-karpathy-skills +59,162★ · anthropics/skills +49,058★ · Shubhamsaboo/awesome-llm-apps +38,266★ · farion1231/cc-switch +36,887★ · nextlevelbuilder/ui-ux-pro-max-skill +35,298★ · Graphify-Labs/graphify +32,474★
风险分级灰犀牛 / 黑天鹅
灰犀牛 · 可预见
  • 灰犀牛Ⅰ FOMC 9/16 加息(约 88%)引发 AI 高估值资产回撤|概率 ≥70%|依据:利率期货定价(深度版风险分级)
  • 灰犀牛Ⅱ 智能体安全失控扩散:48% 已部署智能体无有效安全控制,RubyDoc 类滥用再发|概率 ≥70%|依据:HN 9/12 + 行业报告 + 外滩大会报告
  • 灰犀牛Ⅲ 基准公信力危机升级:菲尔兹奖得主联署后实验室若不改,学术共同体反噬加剧|概率 40-69%|依据:25 位菲尔兹奖得主联署
黑天鹅 · 低概率大冲击
  • 黑天鹅Ⅰ 10/1 OpenAI vs Apple 听证会出现监管转向信号,动摇头部估值锚|概率 <40%|依据:听证会日程
  • 黑天鹅Ⅱ 某头部实验室被曝安全短板,「第三方评估开放」承诺反成照妖镜,引发行业级信任事件|概率 <40%|依据:第三方评估机制推进中
行动建议3
1. 智能体治理套件——切入最急

给已部署智能体的中大型企业(电商/金融客服优先)做「身份认证 + 权限分级 + 行为审计」轻量 SaaS,以 RubyDoc 滥用事件为 demo 案例、30 天免费 POC 获客;进阶按智能体数量收费、接入企业 IAM 做留存;协同智能体框架/模型厂商渠道分佣,借 harness 话题做安全评测内容营销。

时间窗 舆情热度期 3-6 个月,48% 缺口是现成销售话术
2. 算力供应链分包——确定性最高

给戴尔/HPE 国产化承接商、液冷与网络设备厂商做二供/运维交付分包,盯央企集采(单日近 9 亿)的配套服务缺口;路径=运维分包 → 备件与液冷改造 → 进下一轮集采短名单;组合北京生物医药算力专项贴息,用电信 Token/智能体预测做客户 ROI 测算工具。

时间窗 FY2027 CapEx 落地前 12 个月,2026 年内签框架最关键
3. 评测 harness 第三方服务——壁垒最高

给国产模型厂商做 Real-SWE 口径的合规评测报告,助其上私有库口径新榜;路径=单模型报告 → 行业基准订阅 → 承接「放缓联盟」式第三方评估;与行动①共用企业客户渠道,评测报告反哺安全套件销售。

时间窗 harness 成为排名前提的窗口 6-12 个月
未来 3 个月时间线概率
时间窗事件概率依据
9/14Siri AI 如期上线;Claude Code 限额下调生效;DeepSeek V4 Pro 保留原入口高 75%官方公告 + 前瞻日历
9/16FOMC 加息 25bp,AI 资产短期承压高 85-90%利率期货定价
9月内戴尔/HPE/超微股价随订单簿继续走强中 60%900-950 亿 CapEx 点名承接
9/29OpenAI DevDay 发布智能体商业化新框架;上海公安 1212.5 万平台开标高 70%历年惯例 + 已公告
9/30加州法案签署窗口关闭;ARC-AGI-3 截榜中 65%法定窗口
10/1OpenAI vs Apple 听证会聚焦反垄断,无即时结论中 65%听证会属性
10月中旬Anthropic IPO 推进披露(估值冲 2 万亿);「放缓联盟」再纳新成员中 55%路透证实窗口
11月国产模型成本战传导,推理占比数据首次佐证电信报告;国产算力四季度集采收官(央企订单破百亿)中 60-65%成本战传导周期 + 集采节律外推
素材来源:报道台账(滚动查重):2026-09-07 ~ 2026-09-13 共 99 条主题记录 · AI 官方视频监控台账:7 天 · 405 条事件(精选 14 条带链接) · 站点榜单快照:2026-09-03 → 2026-09-10 星标涨幅 Top8 · 风险/行动/时间线:承接 2026-09-13 日报
SkillHub.news — 全球 Skill 榜单
WEEK 2026-09-13 · 46 STORIES · SKILLHUB WEEKLY
SkillHub.news Weekly AI
Sep 7 ~ Sep 13, 2026 · Weekly · sources 4
Highlights46 stories · approx.18min read
01
WEEKLY THREADS24 stories
DeepSeek reverses V4 Pro shutdown (new inflection point; reversal switches on 9/12)
02
OFFICIAL EVENTS14 stories
Quoting Paul Ford
03
RANKING MOVES8 stories
obra/superpowers stars +79,301
01WEEKLY THREADS24 stories

DeepSeek reverses V4 Pro shutdown (new inflection point; reversal switches on 9/12)

报道台账2026-09-13
Official late 9/11 email + 9/12 announcement: after 9/14, original API endpoint retained, billing unchanged, no forced routing to Flash; enterprises can choose versions—three days of developer backlash forced reversal of a business decision; for the first time, users gain a say over API lifecycle (official email/AI Model Watch/tokencost all agree, high)

Dario 'We Must Pace the Frontier' + Altman concurs

Reuters等2026-09-13Confidence High
Early 9/12-13 post: three-step plan + Anthropic unilaterally commits to third-party evaluation, permanent employee-level access; Altman hours later "we will do the same", Musk also weighs in; timing delicate—Anthropic races toward an IPO at ~2 trillion valuation, "slowdown" in same frame as largest IPO (Reuters et al., high)

25 Fields Medalists co-sign criticism of hard-problem benchmark

多源2026-09-13Confidence Mid
25 people including Terence Tao, Peter Scholze, and Pierre Deligne signed 'A Severe Misalignment of AI in Mathematics' on 9/11, with about 2500 people following with signatures; the trigger shares the same origin as the 9/9 Navier-Stokes controversy—the dispute over the legitimacy of academic and lab evaluations has gone public (multi-source, medium-high)

Oracle CapEx landing point: Dell, HPE both hit new highs (Update · Day 2)

业绩会/行情2026-09-13Confidence HighProgress · Day 2
On the earnings call, the CFO reiterated FY2027 capex of $90-95 billion and unusually named Dell/HPE as taking on AI racks/liquid cooling/network equipment; on 9/11 Dell +11.98%, market cap broke $360 billion to a new high, HPE +about 10.5-11%, Supermicro order book $60 billion—AI infrastructure has a 'consignee list' for the first time (earnings call/market, high)

Telecom Research Institute Token report: 10亿亿 in 2026

报道台账2026-09-13
AI Infrastructure Development Research Report in the Agent Era (2026): China's annual Token consumption 10亿亿 (10^18) in 2026, exceeding 3500亿亿 by 2030; inference compute share 80% in 2029; global active agents about 8000万个 in 2026, projected 22.16亿个 in 2030 (CCTV cited, China·authoritative single report)

Central SOE computing power centralized procurement: three orders in one day, nearly 900 million

财富号转述待公告核2026-09-13Confidence Mid
9/12: China Great Wall wins State Grid third-phase server contract over 300 million + iSoftStone Computer wins China Mobile component renewal 418 million + Pingzhi Information pre-wins Yunnan intelligent computing about 137 million (Caifuhao relay, pending announcement verification, China); in the same period, Beijing-Tianjin-Hebei "Space Computing Power Corridor" initiative + Beijing over 100 million biomedical computing power special program (to empower 100 projects in one year)

Real-SWE private codebase benchmark first leaderboard

多源2026-09-13Confidence Mid
All tasks come from authorized private production repositories (billing/tax/migration; reference solutions change a median of 11 files); Fable5.1+Claude Code 38.8%/GPT-6 Astra+Codex CLI 33.8%/Gemini3.8 31.2%/GLM5.3 28.8% — scores are tied to model+toolchain, harness is a prerequisite for ranking (multi-source, medium-high)

SREGym: SRE agents fall 20% short of closing the loop

单源2026-09-13Confidence Mid
GPT-5.6 Sol(max)+Codex tops with 81.0% end-to-end resolution (diagnosis 95.2%/mitigation 85.7%); Claude Opus 5+Claude Code 76.2%; best configuration cannot close the loop on about 20% of production failures, with Token costs differing by nearly 2x across configurations (single-source, medium)

Agent governance whitespace (business opportunity)

三源2026-09-13Confidence High
OpenAI agents abusing RubyDoc caused RubyGems to suspend registrations for 4 days (HN 9/12); 48% of deployed agents lack effective security controls; Bund Summit report deems governance a 'necessary condition'—identity authentication/permission management/behavior auditing SaaS is a China-U.S. common whitespace (three sources, high).

Capital moves downstream: 'water sellers' make three consecutive investments

报道台账2026-09-13Confidence Mid
Enfabrica $125 million (NVIDIA strategic co-investor; interconnect chips)/Yezhixin, one year after founding, cumulative nearly RMB 1 billion (on-device NPU, co-invested by Shenzhen Capital Group and others, team from Apple ANE)/Yuedian Technology FDE tens of millions Pre-A (self-reported nearly 100 customers, break-even, company figures); Mecka AI $500 million valuation single-source, deferred (STAR Market Daily, high/medium)

Nine departments' intelligent driving "15th Five-Year Plan"

工信部等2026-09-13Confidence High
9/9 issued, 9/11 announced: 2030 scaled application of autonomous driving + high-level autonomous driving on highways and urban expressways; sets hard target that "safety performance of vehicles equipped with autonomous driving systems greatly exceeds human drivers"; accelerate mandatory national standards for autonomous driving/automated parking safety—intelligent-driving safety for the first time written as an industrial-planning governance hard target (MIIT et al., high confidence)

Forward calendar update (dense September window)

多源2026-09-13Confidence High
9/14 Siri AI launches with iOS 27 beta + Claude Code weekly limit cut by 17% takes effect; 9/29 OpenAI DevDay + Shanghai police 1212.5万 tender opening; 9/30 California bill window + ARC-AGI-3 leaderboard cutoff; 10/1 OpenAI vs Apple hearing; mid-October Anthropic IPO roadshow (multiple sources, high confidence)

Excluded items archive

报道台账2026-09-13Confidence Mid
Qwen3-Next-80B open-source is rehashed old news from Sep 2025 ('Alibaba US stock +8%' claim is false; actual +0.68%); removed; Zhipu GLM's '9/12 global opening' date wording is wrong (GLM-5.3 was released 8/14 and open-sourced 8/28); NVIDIA's 10 billion investment in Anthropic duplicates the 9/12 'planned 10 billion cornerstone'; Zhipu ARR 1.6 billion is the already-reported 9/5 figure; MiniMax ARR 800 million has no source; US stocks ending a four-day losing streak already reported 9/12; grok-4.6 Fullstack topping the list is only a mirror-site single source and not accepted; Mecka AI valuation single-source, hold; Copyright Center portrait/voiceprint rules / Canada compliance guide only podcast retellings, hold for Monday verification

PaperCut vulnerability weaponized by AI agent (395 organizations/48 countries)

GreyNoise/Blackpoint2026-09-12Confidence High
GreyNoise+Blackpoint dual sources: used CVE-2026-81578/82078 to let Codex(harness)+DeepSeek model agents automate development and execution of attacks; breached 395 organizations across 48 countries in hours, fastest domain admin in 7 minutes (12 of 440 instances full domain admin, cited with caveats)—first at-scale real-world case of agent weaponization (GreyNoise/Blackpoint, high)

Pentagon plans to lend $5 billion to Fluidstack

WSJ/Reuters2026-09-12Confidence High
WSJ 9/10 first reported + Reuters confirmed: via Office of Strategic Capital (OSC), about $5 billion loan to cloud startup Fluidstack for data center component supply chain (not new construction), talks not finalized—AI infrastructure sees first direct state-credit backstop financing (WSJ/Reuters, high)

OpenAI pauses new Pro 20X subscriptions

OpenAI官宣/PCMag2026-09-12Confidence High
Product lead Sottiaux announced on X on 9/10: demand for GPT-6 Astra 'unprecedented'; $200/month 20X tier stops new subscriptions and upgrades, existing subscriptions unaffected—flagship subscription overwhelmed by its own demand, corroborated by token ledger (daily inference cost >$600) (OpenAI official/PCMag, high)

DeepSeek 9/14 flagship retirement-style switch + HBM demand 1/4 (update · Day 2)

DeepSeek官方+行情2026-09-12Confidence HighProgress · Day 2
Official confirmation: from 9/14, V4 Pro fully routed to V4.1 Flash and priced as Flash (model card Terminal-Bench 4.0=31.2); new architecture KV cache needs only 1/4 the HBM and 1/8 the SSD of the old version; SK Hynix US shares -5.2%/Micron -4.9%—algorithmic efficiency for first time pushes back against memory price-hike narrative; AA 39.55/LiveBench 77.3 not independently verified, not adopted (DeepSeek official + market, high)

Moonshot AI ARR breaks 1B, targets 2B (Progress · Day 1)

Bloomberg2026-09-12Confidence HighProgress · Day 1
Bloomberg 9/11: Kimi K3 drives August ARR to $1B (June $300M, +2.3x in 5 weeks), targeting $2B by year-end and advancing financing at $50B pre-money valuation — adding revenue-side validation for A+H listing track (Bloomberg, High)

MIIT "AI+Software" Special Action Plan

工信部/新华社2026-09-12Confidence High
Xinfa [2026] No. 209 (issued 9/2, publicly released 9/11): by 2028 cover 20,000 above-scale software enterprises + 100 technical transformation projects + 100 agent benchmark cases + 5 open-source projects, with compute vouchers; first systematic deployment of security review for AI-generated code and agent identity labeling (MIIT/Xinhua, High)

Anthropic IPO moved up to October + OpenAI files confidentially

Reuters/FT转述2026-09-12Confidence Mid
Reuters 9/12 confirms Anthropic roadshow in mid-October, completion before midterm elections, raising up to $100B / valuation about $2T, NVIDIA planned $10B cornerstone; multiple sources confirm OpenAI confidentially submitted S-1 on 6/8 — safety narrative and capital narrative in same frame (Reuters/FT relayed, Medium-High)

August CPI lands: software component +25.4% enters inflation gauge (Progress · Day 2)

BLS/CME2026-09-12Confidence HighProgress · Day 2
Overall 3.4% in line, core MoM +0.3% above expectations (largest since April), probability of a 25bp rate hike on 9/16 about 87%-90% late in the session (CME readings at different times); CPI component "consumer software and accessories" +25.4% YoY, record high; US stocks closed higher after bad news landed, ending four straight declines (BLS/CME, High)

California SB 867 bans AI companion toys + ASI bill penalties

加州官方/fact sheet2026-09-12Confidence High
Newsom signs SB 867: first in the US to ban AI companion chat features in toys for children under 16 (5 years, no disclosure exemption path); Sanders' 'Ban ASI Act' framework: corporate dissolution-style 'death penalty' + up to 20 years for individuals (9/3 proposal, 9/16 briefing; details subject to official text) (California official/fact sheet, high)

Ant sets tone for agentic commerce moment + Emergence Award

第一财经2026-09-12Confidence High
Han Xinyi at 9/11 Bund Conference: 'agentic commerce ChatGPT moment,' called out 'merchants lack good agent services'; Alipay's annual RMB 10 million 'Agent Emergence Award' (up to RMB 1 million per project) — AEO (agent-callable transformation) is white space (Yicai, high)

Evaluation paradigm shifts to harness: HarnessDev + 方升-Code 2.0

报道台账2026-09-12
ByteDance Seed and others release HarnessDev: same weights, swapping harness yields nearly 15-point score gap (GPT-5 Terminus 2 35.2% vs Codex CLI 49.6%); CAICT 方升-Code 2.0 (5000+ real project data, new engineering quality/stability dimensions) (encorp paraphrase/Tencent News, medium/high)
02OFFICIAL EVENTS14 stories

Quoting Paul Ford

官方·Simon Willison2026-09-13Confidence High
Original ↗

Introducing Music v2.5

YouTube·ElevenLabs2026-09-12Confidence High
Original ↗

Quoting huggingface.co/security.txt

官方·Simon Willison2026-09-12Confidence High
Original ↗

Soft-deprecating re.match()

官方·Simon Willison2026-09-12Confidence High
Original ↗

GPT-Live-1 is now in the API

YouTube·OpenAI2026-09-12Confidence High
Original ↗

Quoting Boris Cherny

官方·Simon Willison2026-09-12Confidence High
Original ↗
03RANKING MOVES8 stories

obra/superpowers stars +79,301

GitHub2026-09-03 → 2026-09-10
Original ↗

mattpocock/skills stars +71,905

GitHub2026-09-03 → 2026-09-10
Original ↗

anthropics/skills stars +49,058

GitHub2026-09-03 → 2026-09-10
Original ↗

farion1231/cc-switch stars +36,887

GitHub2026-09-03 → 2026-09-10
Original ↗
Ecosystem Pulseself-updating
2026-09-03 → 2026-09-10 Star growth:obra/superpowers +79,301★ · mattpocock/skills +71,905★ · multica-ai/andrej-karpathy-skills +59,162★ · anthropics/skills +49,058★ · Shubhamsaboo/awesome-llm-apps +38,266★ · farion1231/cc-switch +36,887★ · nextlevelbuilder/ui-ux-pro-max-skill +35,298★ · Graphify-Labs/graphify +32,474★
Risk Levelsgray rhino / black swan
Gray rhino · foreseeable
  • Gray Rhino I: FOMC 9/16 rate hike (about 88%) triggers AI high-valuation asset pullback | Probability ≥70% | Basis: interest rate futures pricing (deep version risk grading)
  • Gray Rhino II: agent safety loss-of-control spreads: 48% of deployed agents have no effective security controls; RubyDoc-type abuse recurs | Probability ≥70% | Basis: HN 9/12 + industry report + Bund Conference report
  • Gray Rhino III Benchmark credibility crisis escalates: if labs do not change after Fields Medal winners' joint letter, academic community backlash will intensify | Probability 40-69% | Basis: joint letter co-signed by 25 Fields Medal winners
Black swan · low-probability
  • Black Swan I 10/1 OpenAI vs Apple hearing shows regulatory shift signal, shaking top-tier valuation anchor | Probability <40% | Basis: hearing schedule
  • Black Swan II A leading lab exposed for security weaknesses; 'open third-party evaluation' pledge becomes a revealing mirror, triggering an industry-level trust incident | Probability <40% | Basis: third-party evaluation mechanism in progress
Action Items3
1. Agent governance suite — most urgent entry point

For mid-to-large enterprises that have deployed agents (e-commerce/financial customer service first), build lightweight SaaS for 'identity authentication + permission tiering + behavior auditing'; use the RubyDoc abuse incident as a demo case and acquire customers with a 30-day free POC; advanced tier charges by number of agents and integrates with enterprise IAM for retention; partner with agent framework/model vendors on channel commissions, and use the harness topic for security evaluation content marketing.

window Buzz window: 3-6 months; the 48% gap is a ready-made sales pitch.
2. Compute supply chain subcontracting — highest certainty

Provide second-source/O&M delivery subcontracting for Dell/HPE localization contractors and liquid cooling and network equipment vendors; target the supporting-service gap in central SOE procurement (nearly 900 million in a single day); path = O&M subcontracting → spare parts and liquid cooling retrofit → enter next round of procurement shortlist; combine with Beijing biomedicine compute special interest subsidy, and use telecom Token/agent forecasts to build a customer ROI calculation tool.

window With 12 months until FY2027 CapEx deployment, signing a framework within 2026 is most critical.
3. Third-party evaluation harness services — highest barrier.

Produce Real-SWE-standard compliance evaluation reports for domestic model vendors, helping them onto a new private-repo-standard leaderboard; path = single-model report → industry benchmark subscription → take on Slowdown Alliance-style third-party assessments; share enterprise customer channels with Action ①; evaluation reports feed back into security suite sales.

window 6-12 month window for harness to become a prerequisite for rankings.
Next 3 MonthsProb.
WindowEventProb.Basis
9/14Siri AI launches on schedule; Claude Code usage limit cut takes effect; DeepSeek V4 Pro retains original entry point.High 75%Official announcements + forward calendar.
9/16FOMC hikes rates 25bp; AI assets under short-term pressure.High 85-90%Interest rate futures pricing
9月内Dell/HPE/Supermicro shares continue to strengthen with order bookMedium 60%90-95 billion CapEx named to be undertaken
9/29OpenAI DevDay launches new framework for agent commercialization; Shanghai Public Security opens bids for 12.125 million platformHigh 70%Historical precedent + already announced
9/30California bill signing window closes; ARC-AGI-3 leaderboard cutoffMedium 65%Statutory window
10/1OpenAI vs Apple hearing focuses on antitrust, no immediate conclusionMedium 65%Hearing nature
10月中旬Anthropic IPO advances disclosure (valuation surges toward 2 trillion); Slowdown Coalition adds another new memberMedium 55%Reuters confirms window
11月Domestic model cost war spills over; inference share data first corroborates telecom report; domestic compute Q4 centralized procurement wraps up (central SOE orders exceed 10 billion)Medium 60-65%Cost-war transmission cycle + centralized procurement cadence extrapolation
Sources:Coverage ledger (rolling deduplication):2026-09-07 ~ 2026-09-13: 99 topic records total · AI Official Video Monitoring Log:7 days · 405 events (14 selected with links) · Site leaderboard snapshot:2026-09-03 → 2026-09-10 Top8 Star Growth · Risk/Action/Timeline:Follows on from the 2026-09-13 daily report
SkillHub.news — Global Skill rankings