Reports
Report 2026-06-25 160 sources Input Tokens: 93,374 · Output Tokens: 3,549 · Total Tokens: 100,575

AI 戰情報告 (2026-06-25)

openaianthropicgooglemetadatabricksqualcommmistralslack claude-taggemini-3.5-flashjalapeñonova-2-sonicfugu agentchipmcphardwaresecurityautomationvoice-aitoken-rationing

AI 戰情報告 (2026-06-25)

昨日 AI 重點新聞

OpenAI 發表首款自研推理晶片「Jalapeño」 ref-88, ref-157, ref-0, ref-67, ref-153

OpenAI 與 Broadcom 合作,揭曉了名為 Jalapeño 的 ASIC 推理晶片。該晶片在極短的九個月開發週期內完成,專為 ChatGPT、Codex 與未來 AI 代理產品的推理任務設計,旨在降低計算成本並提升垂直整合度。

Anthropic 推出 Claude Tag 代理加入 Slack 協作 ref-2

Anthropic 發表新工具 Claude Tag,正式將 AI 代理的能力帶入 Slack 頻道。團隊成員現在可以像標記同事一樣標記 Claude,讓它主動處理從工程除錯到行銷方案等異步任務。

Qualcomm 斥資 40 億美元收購 AI 晶片軟體公司 Modular ref-142, ref-77, ref-14, ref-5

Qualcomm 確認收購由 LLVM 創辦人領導的熱門新創 Modular。此收購案金額接近 40 億美元,旨在整合 Modular 的高性能 AI 程式語言與推理引擎,以在資料中心與邊緣設備領域挑戰 NVIDIA 的霸權。

Google Gemini 3.5 Flash 推出「電腦操控」功能 ref-11, ref-4

Google DeepMind 宣佈為其輕量級模型 Gemini 3.5 Flash 導入「Computer Use」能力。此更新讓模型能夠像人類一樣透過視覺理解並操作電腦界面,顯著提升了自動化任務處理的範圍。

Databricks 發表 Omnigent 元編排框架與 Genie One 助手 ref-77, ref-76

Databricks 發表開源系統 Omnigent,這是一種用於組合、控制與共享 AI 代理的「元線程裝置 (Meta-harness)」。同時推出 Genie One 代理同事,協助團隊在 Data Lakehouse 環境中實現數據驅動的自動化。

Mistral AI 升級連接器並推出 MCP 除錯工具 ref-2

Mistral AI 推出多項連接器更新,包括加強 Admin 控制權、支援 MCP (Model Context Protocol) 除錯工具以及多帳戶連動。這讓 AI 代理在生產環境中存取 CRM 與代碼庫時更具安全感與透明度。

Meta 加速 AI 審核機制,預計年底取代 90% 人力 ref-8

Meta 內部消息指出,公司正大幅加速使用 LLM 來審核社群內容與廣告。目標在 2026 年底前讓 AI 接管 90% 的人工審核請求,以應對日益成長的平台監管壓力。

Assort Health 獲 1.2 億美元融資發展醫療語音代理 ref-47, ref-127

專注於醫療保健 AI 語音代理的新創公司 Assort Health 完成 C 輪 1.2 億美元融資,估值達 12 億美元。該公司產品能自動處理病患預約、身分核驗等語音交互,解決醫療體系的人力缺口。

Hang Ten Systems 成立:利用 AI 降低企業軟體成本 ref-147, ref-28, ref-57

由前 Infosys 執行長 Vishal Sikka 創立的 Hang Ten Systems 獲得 3200 萬美元種子輪融資。該公司旨在挑戰傳統 IT 服務外包模式,利用 AI 代理來協助企業以更低成本維護與自定義企業級軟體。

Grab 與 Snyk 針對 AI 代理推出安全執行與治理平台 ref-141, ref-61

Grab 開發了 Kubernetes 原生安全平台 Palana 以隔離 AI 代理的不可預測行為;Snyk 則推出 Evo Agentic Security,專門針對 AI 編程代理的執行權限進行安全治理。

Sakana AI 推出 Fugu 多代理編排系統 ref-2

日本 AI 研發公司 Sakana AI 發表 Fugu,這是一個能動態拆解任務並路由至不同專業模型(如 Claude 或 OpenAI)的編排系統,宣稱能以「集體智慧」達到旗艦模型的性能。

Hootsuite 與 Facebook 為創作者推出專屬 AI 助手 ref-34, ref-54, ref-84

Facebook 重新推出 Creator Studio 獨立 App 並內建 AI 聊天機器人協助成長;Hootsuite 則推出 Wisdom 行銷代理,能將社交訊號直接轉化為自動化的行銷行動建議。

Notion 整合 Cursor SDK 嵌入 AI 編程代理能力 ref-17

Notion 宣佈採用 Cursor SDK,在其平台上直接嵌入具備「環境感知」的編程代理。這讓非專業開發者也能透過對話來調整 Notion 的內部功能與自動化流程。

MIT 與微軟合作開發 Murakkab 系統優化代理工作流 ref-121

MIT 與 Microsoft Azure 共同研發 Murakkab 智能系統,能自動根據用戶需求優化 Agentic Workflow 的模型選擇與硬體配置,大幅提升執行速度並降低能耗與成本。

Gartner 警示:企業因預算壓力開始實施「Token 配給」 ref-8, ref-33

Gartner 與相關研究指出,由於員工頻繁使用 AI 處理瑣碎任務導致預算超支,企業界正進入「Token 配給」時代。預測到 2028 年,AI 編程的成本甚至可能高於直接聘僱開發人員。

市場洞察

1. 代理編排層的標準化競爭進入白熱化

延續昨日關於 Sakana AI Fugu (ref-132) 挑戰旗艦模型的報導,今日 Databricks 發表 Omnigent (ref-4, ref-105) 展示了市場對「元編排 (Meta-harness)」的強烈需求。當企業內部出現多樣化的代理工具時,如何建立一個跨模型、跨平台的通用 API 與安全控制層,已成為繼模型本身之後最關鍵的戰場。這也解釋了為何 Mistral 積極推動其連接器與 MCP 除錯工具 (ref-9)。

2. 垂直領域 AI 語音代理的商業化突破

今日 Assort Health 的高額融資 (ref-10) 與 AWS Nova 2 Sonic 的醫療實例 (ref-40, ref-67) 顯示,Voice AI 在高合規要求的醫療行業已具備成熟的生產力價值。這種結合身分核驗與後端系統自動對接的「語音 + 代理」模式,正迅速取代傳統的語音熱線,這與昨日報告中提到的「代理化」替代人力趨勢高度吻合。

3. AI 硬體整合的深水區:從運算力轉向推理經濟學

OpenAI 發表 Jalapeño 晶片 (ref-4, ref-48) 與 Qualcomm 收購 Modular (ref-41) 反映出一個重要趨勢:AI 巨頭正致力於優化「推理 (Inference)」而非僅是訓練。隨著企業開始進行 Token 配給 (ref-63, ref-79),降低單位推理成本與提升單位能效 (ref-6) 成為 AI 產品能否具備財務可持續性的核心護城河。

Sources 160 items
ref-0 Hacker News OpenAI unveils its first custom chip, built by Broadcom ref-1 Hacker News RubyLLM: A Ruby framework for all major AI providers ref-2 Hacker News Anthropic says Alibaba illicitly extracted Claude AI model capabilities ref-3 Hacker News PR spam today looks like email spam in the early 2000s ref-4 Hacker News Computer use in Gemini 3.5 Flash ref-5 Hacker News Qualcomm to Acquire Modular ref-6 Hacker News Why eval startups fail (2025) ref-7 Hacker News Markdy: Like Mermaid Diagrams, but for Motion ref-8 Hacker News What I'm Finding About LLM Code Style and Token Costs ref-9 Hacker News Show HN: Wordit – Change One Letter, Keep the Chain Going ref-10 Hacker News Wikipedia Workers in Britain set global first by seeking union recognition ref-11 Google DeepMind Introducing computer use in Gemini 3.5 Flash ref-12 Google Developers Blog Unlocking the Power of the TPU Stack: Introducing our new Developer Hub ref-13 Google Developers Blog Enhance Security and Trust: New Session Metadata in Sign in with Google ref-14 Hugging Face Blog Accelerating Transformers Fine-Tuning with NVIDIA NeMo AutoModel ref-15 Mistral AI News Bringing more control over your connectors ref-16 xAI News Explore the markets with Interactive Brokers and Grok ref-17 Cursor Blog How Notion used the Cursor SDK to embed coding agents ref-18 AWS ML Blog Huntington Bank: Redacting sensitive data from 400M+ documents with AWS ref-19 AWS ML Blog Build a healthcare appointment agent with Amazon Nova 2 Sonic ref-20 AWS ML Blog AI-powered BI with Snowflake and Amazon Quick ref-21 AWS ML Blog How Loka Built a Natural, Low-Latency Voice Agent with Amazon Nova 2 Sonic ref-22 The Rundown AI Meet your new Slack coworker — Claude ref-23 Latent Space [AINews] It's Meta-Harness Summer ref-24 Latent Space Why the Frontier Ecosystem must be Open — Matei Zaharia and Reynold Xin, Databricks ref-25 Simon Willison simonw/browser-compat-db ref-26 Simon Willison Quoting Tom MacWright ref-27 TechCrunch AI Europe is pushing back on Washington’s chip war ref-28 TechCrunch AI Former Infosys chief has a new startup that wants to challenge the IT services world ref-29 TechCrunch AI Cerebras stock plunges after earnings as CEO says margin outlook was misunderstood ref-30 TechCrunch AI AI was supposed to kill engineering jobs, but new data suggests they’re the most resilient ref-31 TechCrunch AI AI researchers continue to leave Google for its rivals ref-32 TechCrunch AI The memory chip crunch is paying off for this US company ref-33 TechCrunch AI Companies are scrambling to stop employees from maxing out AI budgets with small tasks ref-34 TechCrunch AI Facebook rolls out an AI companion app for creators ref-35 TechCrunch AI Agility Robotics plans to go public via SPAC in a $2.5B deal ref-36 TechCrunch AI Figma adds code layers, support for animations, more AI features in new update ref-37 TechCrunch AI 3 days left to save up to $190 on your TechCrunch Founder Summit 2026 pass ref-38 The Verge AI Congresswoman denies staff used AI to write defense funding amendment ref-39 The Verge AI The $27 million Al proxy war over Alex Bores ends in a draw ref-40 The Verge AI Figma now has AI motion graphics and shader tools ref-41 The Verge AI OpenAI reveals its first AI processor: Jalapeño ref-42 The Verge AI The Google Home Speaker sounds good and looks great — but it’s finicky ref-43 Ars Technica One-two punch delivered in global operation disrupts cybercrime "assembly line" ref-44 MIT Technology Review AI The emergence of the web data infrastructure layer for AI ref-45 TechMeme As China's working-age population shrinks, a consensus is emerging that it must deploy embodied AI robots into as many tasks as possible, as soon as possible (Financial Times) ref-46 TechMeme How a $45M donation to a group supporting Trump brought Larry Ellison deeper into Trump's circle, helped Oracle's business, and his son's acquisition of WBD (Wall Street Journal) ref-47 TechMeme Assort Health, which develops AI voice agents for health care to handle scheduling and more, raised a $120M Series C led by Menlo Ventures at a $1.2B valuation (Heather Landi/Fierce Healthcare) ref-48 TechMeme Paris-based Tsuga, whose observability software runs in customers' own clouds to help them skip per-byte ingestion costs, raised a $35M Series A led by Singular (Ana Maria Constantin/The Next Web) ref-49 TechMeme TRM Labs: Iran-linked wallets, including those tied to Iran's central bank, have moved over $3.84B in transactions through the crypto exchange CoinEx since 2019 (Wall Street Journal) ref-50 TechMeme NAND flash maker Kioxia, Japan's most valuable company since June 12, plans to offer US depositary shares in spring 2027 amid AI-driven demand for memory chips (Bloomberg) ref-51 TechMeme Sources: Meta is accelerating plans to use LLMs to review content and ads, replacing ~50% of human review requests in 2026 and aiming for 90% by year-end (Financial Times) ref-52 TechMeme New Zealand-based Partly, which develops an AI model for the auto parts industry, raised a $50M Series B led by DST at a $500M valuation, as it enters the US (Duncan Riley/SiliconANGLE) ref-53 TechMeme The Netherlands is lobbying the US not to expand chip equipment export controls that would constrain ASML's ability to sell immersion DUV machines to China (Bloomberg) ref-54 TechMeme Facebook brings back Facebook Creator Studio as a stand-alone app with built-in AI chatbot to help creators grow their audiences through personalized guidance (Aisha Malik/TechCrunch) ref-55 TechMeme Digital advocacy firms like CiviClick and Influent appear to use AI to generate mass public comments on local energy projects, mostly favoring fossil fuel use (Bloomberg) ref-56 TechMeme An Amazon seller reveals how middlemen on chat apps offer access to Amazon employees who allegedly grant favors, like reinstating suspended accounts, for a fee (Spencer Soper/Bloomberg) ref-57 TechMeme Hang Ten Systems, led by former Infosys CEO Vishal Sikka, launches with a $32M seed led by Mayfield to help enterprises use AI to run software at a lower cost (Jagmeet Singh/TechCrunch) ref-58 TechMeme Trase, which is building an operating system and infrastructure layer for AI agents in industries like health care and defense, raised a $107M seed led by Arch (Brock E.W. Turner/Axios) ref-59 TechMeme Paris-based digital health insurance startup Alan raised €480M led by Prosus at a €5.5B valuation, and is set for €1B+ in annual recurring revenue by year's end (Tim Bradshaw/Financial Times) ref-60 InfoQ AI/ML Slack Outlines Four-Phase Journey to a Multi-Cloud AI Serving Platform ref-61 InfoQ AI/ML Grab Builds Secure Agentic AI Workload Platform ref-62 InfoQ AI/ML Anthropic Lead: HTML Increasingly Better Than Markdown at Keeping Humans Engaged in Agentic Loops ref-63 InfoQ AI/ML Google OpenRL is an Experimental Self-hosted API for LLM Post-Training Fine-tuning ref-64 InfoQ AI/ML AI Is Moving up the Software Lifecycle: From Code Review to PRD Governance ref-65 InfoQ AI/ML Presentation: Rules for Understanding Language Models ref-66 InfoQ AI/ML Article: Beyond CLEAN and MVP: Architecting an Offline-first Reactive Data Layer in Android ref-67 The New Stack OpenAI wants to claim more of the AI stack with Jalapeño, its first custom chip ref-68 The New Stack Azul wants to find your unpatched JVMs before AI does ref-69 The New Stack Will it Mythos? One coder’s verdict on Anthropic’s blend of debugging ref-70 The New Stack Chainguard targets Java’s unpatched vulnerability backlog with drop-in remediated libraries ref-71 The New Stack Agentic infrastructure operations begin with accurate, reliable infrastructure data ref-72 The New Stack Sakana Fugu is more than a router. But it’s not the blueprint for AI sovereignty, either. ref-73 Wired AI How to Opt Out of Google Search’s New AI Data Training Feature ref-74 Wired AI A24 Knows You’re Mad About the Google AI Collab ref-75 Wired AI I Met With China’s Top AI Experts. They’re Freaking Out, Too ref-76 Wired AI The Trump White House Is Over Anthropic CEO Dario Amodei ref-77 Wired AI Qualcomm Buys Buzzy Chip Startup Modular for Nearly $4 Billion ref-78 TechCrunch Startups Here’s why Slate changed the battery in its cheap EV truck ref-79 TechCrunch Startups Valor Equity Partners looks to raise a $2.5B Fund VII, per Bloomberg ref-80 TechCrunch Startups Slate Auto’s radically simple electric truck starts at $24,950 ref-81 Crunchbase News The Crunchbase Tech Layoffs Tracker ref-82 Crunchbase News Exclusive: XCures Lands $46M Series B To Clean Up Messy Medical Records With AI ref-83 Crunchbase News Why Ex-Meta CTO Mike Schroepfer Says It’s A Great Time To Build A Hard Tech Company: ‘Infrastructure Is The Moat’ ref-84 Stratechery My Vibe Coding Adventure, The App and the Experience, Ten Takeaways ref-85 Tom's Hardware Best Amazon Prime Day tech deals live on day three — PC hardware deals on GPUs, CPUs, SSDs, and more ref-86 Tom's Hardware Grand Theft Auto 6 preorders begin tonight at midnight local time in the US; here's where to buy — get yours now, it's in the garage and ready to roll ref-87 Tom's Hardware Prime Day brings huge savings on affordable 3D printers — top value picks from Anycubic, Bambu Lab, Elegoo, and Creality hit rock-bottom pricing, can’t-miss deals on filament bundles ref-88 Tom's Hardware Broadcom and OpenAI unveil custom-built Jalapeño inference processor — OpenAI's first chip is a massive reticle-sized ASIC built in an ultra-fast nine-month development cycle ref-89 Tom's Hardware You can still build a great $1000 budget gaming PC with Amazon Prime Day parts — 32GB of RAM and RTX 5060 Ti power beats out the Steam Machine and cheap prebuilts ref-90 Tom's Hardware Some of the best PC speakers we've tested are on sale for Prime Day — save up to 36% on Onkyo, Edifier, and Audioengine speakers ref-91 Tom's Hardware US Secures Netherlands for Pax Silica Alliance in key win for strategic chip alliance — tension remains over MATCH Act restrictions ref-92 Tom's Hardware China tops the list of fastest supercomputers with a CPU-only behemoth, ending US champion El Capitan's reign — 2.198 exaflops of performance without a single GPU ref-93 Tom's Hardware Pay just $149.99 for the TP-Link Archer Wi-Fi 7 router with 9.3 Gbps of bandwidth, now 40 percent off — high-powered BE550 router comes with a full complement of 2.5 Gbps LAN ports, too ref-94 Tom's Hardware Secretlab gaming chairs and desks hit Prime Day week sales, up to $129 off — save on the Titan Evo, Magnus, and more ref-95 Tom's Hardware Geekom Prime Day deals take up to 34% off a new mini PC with our exclusive promo code — Get an AMD or Intel mini PC for less now ref-96 Tom's Hardware These four portable monitors are on sale for Prime Day: Get a second screen for your PC or console for as little as $39 ref-97 Tom's Hardware Get 40% off Samsung's P9 microSD Express card for Nintendo Switch 2 — 512GB of storage just $119 in this limited-time deal ref-98 Tom's Hardware TSMC is reportedly hiking prices for 'all advanced nodes,' accounting for 74% of the company’s wafer business — Nvidia, AMD, Apple, Qualcomm, and others will face higher wafer costs ref-99 Tom's Hardware Logitech goes off the rails with Prime Day gaming mouse savings of up to 47% - Eight of its top-tier gaming mice with incredible discounts ref-100 Tom's Hardware SK hynix files to raise up to $29 billion in historic Nasdaq listing — all proceeds going to advanced AI memory fabs and EUV tool orders ref-101 Tom's Hardware AI data center boom hits a human bottleneck — critical skilled labor shortages could slow deployment despite billions in funding ref-102 Tom's Hardware These Bambu Lab Prime Day week deals are an absolute steal right now, with up to 52% off — big price cuts on new 3D printers, filament, and accessories, including the P1S and A1, starting from $209 ref-103 Tom's Hardware Get an all-AMD 4K gaming PC with Ryzen 7 9800X3D and Radeon RX 9070 XT for just $1,749 — Walmart has slashed $750 off this prebuilt desktop ref-104 Tom's Hardware Valve Steam Machine price hike similar to Steam Deck's 45% increase, company confirms — was probably priced competitively against the PS5 Pro before the RAMpocalypse ref-105 Tom's Hardware Save up to 53% on Lenovo’s Black Friday in June sale — Sitewide discounts on ThinkPads, Legions, and more ref-106 Tom's Hardware Here are the best OLED gaming monitor deals you can snag for Amazon Prime Day — beautiful monitors up to 47% off ref-107 Tom's Hardware China black market Nvidia prices rocket in wake of smuggling crackdown and customs freeze — five-year-old A100 servers triple in price, now fetching up to $82,000 ref-108 Tom's Hardware Netgear rolls out stellar Prime Day deals on Wi-Fi 7 gear — up to 33% off for routers, mesh systems, and range extenders ref-109 Tom's Hardware Get 32GB of Corsair DDR5 RAM for just $214 right now, thanks to these Newegg bundles — AMD and Intel gaming PC build kits massively undercut $350 market prices ref-110 Tom's Hardware This expansive 49-inch OLED monitor is less than $700 in Newegg's Prime Day rival sale — 144Hz 1440p beast is $150 off ref-111 Tom's Hardware Meta pauses mandatory AI training program that tracked employee keystrokes after internal data leak exposed sensitive staff information company-wide — employees express frustration over poor handling of data ref-112 Politico Tech White House helped Mark Zuckerberg and the Google CEO dodge a Senate grilling ref-113 The Hacker News (Security) Cisco Catalyst SD-WAN Zero-Day CVE-2026-20245 Exploited to Gain Root Access ref-114 The Hacker News (Security) CISA Warns Critical Lantronix EDS5000 Flaw Is Being Actively Exploited ref-115 The Hacker News (Security) Amadey and StealC Malware Network Disrupted, 27M Stolen Credentials Recovered ref-116 The Hacker News (Security) Cordyceps CI/CD Flaws Expose 300+ GitHub Repositories to Supply-Chain Attacks ref-117 The Hacker News (Security) Dawn of the Apex Agentic Adversary ref-118 The Hacker News (Security) DoJ Seizes Huione Cloud Account Tied to Cyber Scam Money Laundering ref-119 EU Startups 5 hidden mistakes that are killing your international expansion before it starts ref-120 MIT Research How architecture influences political activity ref-121 MIT Research Improving the speed and energy-efficiency of AI agents ref-122 MIT Research What happens when environmental change outpaces life’s ability to adapt? ref-123 Lobsters Using the Gini Coefficient to Plan Edge Capacity ref-124 Lobsters Unlimited-OCR: One-shot Long-horizon OCR ref-125 Lobsters Flow’s OCaml to Rust Port ref-126 Google News Anthropic Claims Alibaba Ran ‘Brazen’ Campaign to Access Its Claude AI Model ref-127 Google News Assort Health scores $120M series C to scale voice AI agent platform for healthcare ref-128 Google News Linux Foundation prepares open standard for AI agent verification ref-129 Google News Anthropic disables new AI model after White House security directive ref-130 Google News Gartner warns AI model advantage is shrinking ref-131 Google News AI Coding Could Cost More Than Hiring Developers by 2028, Gartner Says ref-132 Google News Hootsuite Rebuilds for the AI Era, Introducing Wisdom — a Social-First AI Agent Turning Social Signals into Action ref-133 Google News Anthropic sends Congress letter accusing Alibaba of illicitly accessing AI models ref-134 Google News Norway Announces Major Ban on Generative AI Use in Elementary School Classrooms ref-135 Google News Google delays Gemini 3.5 Pro launch to July as it tweaks its new frontier AI model ref-136 Google News SpaceX agrees to acquire popular AI coding assistant ref-137 Google News Anthropic AI Model Uncovers Vulnerabilities In Classified U.S. Govt Systems During Testing: Report ref-138 Google News Alibaba slashes Qwen AI model price during US working hours in fight for users ref-139 Google News Google DeepMind Admits Large-Scale AI Agent Deployment Is Unsafe Today ref-140 Google News China's new GLM-5.2 model from Zhipu AI ignites the AI coding race. ref-141 Google News Snyk launches Evo Agentic Development Security to govern AI coding agents | brief | SC Media ref-142 Google News Qualcomm Acquires AI Startup Modular To Challenge NVIDIA's Dominance ref-143 Google News AI startup Engram raises $98 million at $600 million valuation with just 13 employees ref-144 Google News Qualcomm (QCOM) Confirms $3.9B Acquisition of AI Startup Modular ref-145 Google News Qualcomm Stock Forecast: QCOM Jumps 13% After-Hours as $40B AI Revenue Target Stuns Wall Street ref-146 Google News SpaceX secures $6.3bn compute capacity deal from AI startup Reflection ref-147 Google News Former Infosys CEO Vishal Sikka launches AI startup Hang Ten Systems; secures $32 million funding ref-148 Google News HSBC, Google Cloud sign AI partnership, eyeing $100M return per use case ref-149 Google News AI startup Baseten hits $13 billion valuation as Australia's Blackbird makes record bet ref-150 Google News A24 Wouldn’t Tell Hollywood How It Works. Then It Told Google. ref-151 Google News Broadcom's AI Revenue Is on Pace to Triple to $16 Billion in a Single Quarter, but the Stock Just Dipped 17%. Is This an Opportunity to Buy the Dip? ref-152 Google News FRB Releases Open-Source AI Policy on GitHub, Invites Every Industry to Raise the Floor on AI Governance ref-153 Google News Florida sues OpenAI and Altman in landmark AI safety case ref-154 Google News World risks 'losing control of technology' without AI governance, Chinese PM warns ref-155 Google News RBI proposes AI governance framework for banks, mandates human oversight of model driven decisions ref-156 Google News AI PACs pour $20 million into New York Democratic primary in AI regulation battle ref-157 Google News OpenAI and Broadcom Unveil Custom A.I. Chip Design ref-158 Google News China Built The World’s Fastest Supercomputer Without A Single GPU ref-159 Google News Intel-Backed AI Chip Firm SambaNova to Quintuple Valuation to $10 Billion