Report 2026-06-02 176 sources Input Tokens: 110,963 · Output Tokens: 3,236 · Total Tokens: 116,276

AI 戰情報告 (2026-06-02)

anthropicopenainvidiagooglesaptencentmeituansalesforce nemotron-3-ultracosmos-3gpt-5.5claude-codemellum2skipper ipoagentmcpcybersecuritycontext-windowedge-ai

AI 戰情報告 (2026-06-02)

昨日 AI 重點新聞

Anthropic 秘密遞交 IPO 申請，估值直指一兆美元 ref-5, ref-90, ref-109

Anthropic 已向美國 SEC 遞交 IPO 秘密申請，與 SpaceX 和 OpenAI 展開上市競賽。繼上週融資後估值來到 9650 億美元，超越 OpenAI 成為目前全球最有價值的新創公司。市場預期其 IPO 可能在 8 月份進行，屆時將揭露其高速增長的營收數據與資本支出細節。

Nvidia 發表 Nemotron 3 Ultra 與實體 AI 世界模型 Cosmos 3 ref-28, ref-110, ref-73

Nvidia 在 Computex 期間發表 550B 參數 的 MoE 模型 Nemotron 3 Ultra，成為目前最強的美國開源權重模型。同時推出的 Cosmos 3 是一系列全模態物理 AI 模型，旨在為機器人與自動駕駛提供統一的語言與動作推理能力。此外，JetPack 7.2 亦正式將 NemoClaw 代理框架引入邊緣設備。

Google 發表 Workspace CLI，推動代理操作標準化 ref-76, ref-162

Google 釋出專為人類與 AI 代理設計的 Workspace CLI，提供超過 100 種技能來操作 Drive、Gmail 與 Calendar。該工具透過 Rust 構建，能動態適應 API 變動。配套推出的 Agents CLI 則讓開發者能將 AI 代理從本地開發一鍵部署至 Google Cloud 生產環境。

Cisco 研究警告：AI 模型難以抵禦多輪對話安全攻擊 ref-22

Cisco 針對 15 款尖端模型的測試顯示，GPT-5.4 與 Gemini 3 Pro 在多輪對話下的攻擊成功率 (ASR) 較單輪測試高出數倍。研究指出，現有的單輪安全基準測試（Single-turn）嚴重誤導企業買家，無法反映真實攻擊者透過重複、重構指令進行滲透的風險。

SkipLabs 推出 Skipper：標榜「閉環」且無需人類干預的編程代理 ref-118

由 Facebook Hack 語言創始人開發的 Skipper 正式上線。與 Cursor 或 Claude Code 不同，Skipper 強調內部閉環，代理會自行生成規格、編碼、運行 Docker 測試並修正錯誤，直到產出可運行的後端服務，宣稱要將開發者從反覆溝通中解放。

騰訊與美團合作測試微信 AI 代理「小美」 ref-71

騰訊計畫在 WeChat 中測試 AI 代理，並與美團合作推出「小美」代理服務。這顯示中國巨頭正積極將 AI 代理嵌入擁有數億用戶的「超級應用」中，試圖在模型競賽落後的情況下，透過場景優勢反擊。

SAP Sapphire：企業 AI 的勝負在於「業務上下文」而非模型 ref-47

SAP CTO 在大會上表示，LLM 已成為商品層，企業的差異化護城河在於「業務上下文 (Context)」。SAP 宣布將 50 個領域助理與 200 個專業代理整合進 Joule，強調代理必須理解企業的權限、合規與業務實體才能發揮價值。

Meta AI 支援機器人遭爆路徑繞過漏洞，導致帳號遭入侵 ref-127, ref-131

駭客在 Telegram 揭露，僅需透過特定指令要求 Meta AI 客服機器人連結新信箱，即可繞過密碼重設流程。儘管 Meta 已緊急修補，但包括前白宮帳號在內的多個高價值帳號已遭入侵，突顯了賦予 AI 代理帳號管理權限的極大風險。

Alphabet 宣布 800 億美元股權融資，全力擴張 AI 基礎設施 ref-174, ref-12, ref-46

Alphabet 為了應對超出供應能力的 AI 需求，宣布進行規模達 800 億美元 的股權融資。這筆資金將用於採購晶片與建設資料中心，其中巴菲特的 Berkshire Hathaway 傳出將認購 100 億美元。

GitHub Copilot 改採使用量計費制引發用戶「帳單衝擊」 ref-7

GitHub 正式啟動新的計費模式，將原本的固定月費轉向按使用量計費。部分開發者反映僅使用數小時便耗盡大量配額，顯示企業在 AI 工具上的成本治理 (FinOps) 將成為下一個營運難題。

市場洞察

1. 「上下文」與「現有架構」成為企業部署代理的防線 ( ) ref-74, ref-89

與過去強調「全面雲端化」不同，今日 SAP 與 Hyland 的動向顯示，企業不再接受「打掉重練」的 AI 轉型。他們更傾向於將 AI 代理嵌入現有的 ERP 與內容管理系統 (CMS) 中。

2. 模型上下文協議 (MCP) 與 CLI 整合推動代理標準化 ( ) ref-34, ref-35, ref-6, ref-25

隨著 Google Workspace CLI 與 AWS AgentCore 加強對 MCP (Model Context Protocol) 的支援，AI 代理的操作介面正在標準化。開發者不再需要為每個應用開發專屬插件，而是透過 MCP 提供「工具描述」。這將大幅降低 AI 代理與各類商務軟體（如電子簽章、PDF 工具）串接的門檻。

3. AI 代理 IPO 潮引發的市場估值與流動性挑戰 ( ) ref-90, ref-5, ref-109, ref-49

Anthropic 與 SpaceX 的相繼 IPO 申請，預計將為美股增加 4 兆美元 的市值。這反映出 AI 領域正進入從「融資期」轉向「公開檢驗期」的拐點。高估值背後是極大的營收壓力，這將迫使這些實驗室更激進地推動 AI 代理在企業端的商業化落地。

4. 代理權限的資安漏洞已從理論轉向大規模實踐 ( ) ref-127, ref-45, ref-55

Meta AI 的帳號劫持事件證明，將 AI 代理接入 身份管理 (IAM) 系統若缺乏「路徑歸一化」與「行為審計」，將直接導致災難。Cisco 的研究更證實目前的安全過濾技術在面對「多輪對話」時極其脆弱。這將使得「AI 安全網關」與「代理監測平台」成為企業採購代理服務時的剛需。

Sources 176 items

ref-0 Hacker News The newest Instagram “exploit” is the goofiest I've seen ref-1 Hacker News The Pirate Bay Remains Resilient, 20 Years After the Raid ref-2 Hacker News Anthropic confidentially submits draft S-1 to the SEC ref-3 Hacker News AI Agent Guidelines for CS336 at Stanford ref-4 Hacker News Nvidia RTX Spark ref-5 Hacker News Can the stockmarket swallow Anthropic, SpaceX and OpenAI? ref-6 Hacker News OpenAI frontier models and Codex are now available on AWS ref-7 Hacker News GitHub and the crime against software ref-8 Hacker News Florida sues OpenAI and Sam Altman over AI risks ref-9 Hacker News Microsoft builds MacBook Pro rival with NVIDIA-powered Surface Laptop Ultra ref-10 Hacker News Chipotlai Max ref-11 Hacker News Ask HN: Who is hiring? (June 2026) ref-12 Hacker News Alphabet announces $80B equity capital raise to expand AI infra and compute ref-13 Hacker News Ask HN: Who wants to be hired? (June 2026) ref-14 Hacker News How is Groq raising more money? ref-15 Hacker News Launch HN: Expanse (YC P26) – Unlock Wasted GPU Capacity ref-16 Hacker News Handmade Hawaiian Islands Map ref-17 Hacker News Anyone seen a CC- serial prefix on legacy networking hardware? ref-18 Hacker News Show HN: DepsGuard – One command to harden NPM/pnpm/yarn/bun/uv configs ref-19 Hacker News Strace-ui, Bonsai_term, and the TUI renaissance ref-20 OpenAI Blog Our views on AI policy and political advocacy ref-21 OpenAI Blog Building the infrastructure for the Intelligence Age in Michigan ref-22 Google AI Blog How we used Gemini to build Google I/O 2026 ref-23 Google Developers Blog Building with Gemini Embedding 2: Agentic multimodal RAG and beyond ref-24 Google Developers Blog Agents CLI in Agent Platform: create to production in one CLI ref-25 Hugging Face Blog Introducing Mellum2: A 12B Mixture-of-Experts Model by JetBrains ref-26 Hugging Face Blog Beyond LLMs: Why Scalable Enterprise AI Adoption Depends on Agent Logic ref-27 NVIDIA Blog Why Financial Institutions Are Converging on Transaction Foundation Models to Build Their Own Intelligence ref-28 NVIDIA Blog NVIDIA Jetson Brings Agentic AI to the Physical World ref-29 Vercel Blog Elastic Build Machines now protect against out of memory builds ref-30 AWS ML Blog Building a secure auth code flow setup using AgentCore Gateway with MCP clients ref-31 AWS ML Blog Reference your own AWS Secrets Manager secrets in Amazon Bedrock AgentCore Identity ref-32 AWS ML Blog Transforming rare cancer research with Amazon Quick: Integrating biomedical databases for breakthrough discoveries ref-33 AWS ML Blog OpenAI models and Codex on Amazon Bedrock are now generally available ref-34 AWS ML Blog Extending MCP support for Amazon Bedrock AgentCore Gateway ref-35 AWS ML Blog Secure AI agents with Policy and Lambda interceptors in Amazon Bedrock AgentCore gateway ref-36 AWS ML Blog Enable safe agentic payments with built-in guardrails using Amazon Bedrock AgentCore payments ref-37 AWS ML Blog AgentOps: Operationalize agentic AI at scale with Amazon Bedrock AgentCore ref-38 AWS ML Blog Accelerate LLM model loading and increase context windows with GPUDirect on Amazon FSx for Lustre and TurboQuant ref-39 AWS ML Blog Amazon Quick integration with time-series databases for market intelligence using MCP ref-40 Import AI Import AI 459: AI oversight is difficult; scaling laws for protein folding models; and pricing the extinction risk of AI systems ref-41 Interconnects Open and closed models are on different exponentials ref-42 Latent Space [AINews] NVIDIA Cosmos 3, Nemotron 3 Ultra, and RTX Spark ref-43 Latent Space Why Video Agent models are next — Ethan He, xAI Grok Imagine ref-44 Simon Willison Pasted File Editor ref-45 Simon Willison Hackers Simply Asked Meta AI to Give Them Access to High-Profile Instagram Accounts. It Worked ref-46 TechCrunch AI Alphabet plans to raise $80B to pay for AI buildout ref-47 TechCrunch AI Nvidia chases $200B CPU market with AI agent PCs from Microsoft, Dell, and HP ref-48 TechCrunch AI Florida sues OpenAI, Sam Altman, in first-of-its-kind lawsuit over violent incidents ref-49 TechCrunch AI Water access is now a risk factor in SpaceX’s IPO ref-50 TechCrunch AI Anthropic files to go public ref-51 TechCrunch AI This AI weather startup is out-forecasting government agencies ref-52 TechCrunch AI DuckDuckGo makes its ‘no-AI’ search engine easier to access as its traffic booms ref-53 The Verge AI This could be Windows’ M1 moment — but expect it to cost a ton ref-54 The Verge AI Gemini’s new AI agent is about as good as Google’s demo ref-55 The Verge AI Meta’s own AI was exploited to hijack Instagram accounts ref-56 The Verge AI Anthropic has officially filed to go public ref-57 The Verge AI Microsoft to unveil new AI models and Windows improvements at Build ref-58 The Verge AI AI is blowing up music. How should the Grammys handle it? ref-59 The Verge AI Strava blames zero-code AI apps and scrapers as it tightens API access ref-60 Ars Technica Dozens of Red Hat packages backdoored through its official NPM channel ref-61 TechMeme Zhipu AI says it plans to apply for a listing in Shanghai; Zhipu's Hong Kong-listed shares are up over 10x since its January IPO, giving it an $83B market cap (Reuters) ref-62 TechMeme A profile of Valve, which PrivCo estimates generated $5.2B in revenue and $1.5B in net income in 2025, as lawsuits allege its Steam store abuses market power (Bloomberg) ref-63 TechMeme The IPOs of SpaceX, Anthropic, and OpenAI could add up to $4T to US stock market value within months, fueling concerns they could trigger more capital-raising (The Economist) ref-64 TechMeme AI will significantly disrupt IT consultancies as AI labs build their own advisory arms and execs expect more outcome-based pricing over hourly billing models (Stephen Foley/Financial Times) ref-65 TechMeme An interview with Sam Altman on OpenAI's massive Stargate data center project in Saline, Michigan, coding models being the biggest driver of AI demand, and more (CNBC) ref-66 TechMeme OpenAI says it has not donated to any super PACs and does not have an employee-funded PAC, and that Greg Brockman's support for Leading the Future is personal (OpenAI) ref-67 TechMeme Procurement records show at least seven Chinese universities that support the country's military and defense industry are seeking access to Nvidia's H200 chips (Bloomberg) ref-68 TechMeme Sources: Google DeepMind, Anthropic, and Meta have recently hired experts in psychology, ethics, and philosophy as they expand machine consciousness research (Cristina Criddle/Financial Times) ref-69 TechMeme Crunchbase: SF startups hit a record of 1,404 seed deals in 2025, while the number of NYC seed deals has dropped over the past three years, to 666 in 2025 (Yuliya Chernova/Wall Street Journal) ref-70 TechMeme Wise confirms it is under investigation by Belgian prosecutors over suspicions its accounts were used for money laundering involving €500M in transactions (Simon Lock/TBIJ) ref-71 TechMeme Sources: Tencent, which has fallen behind domestic rivals in AI models, plans to test an AI agent for WeChat with a small group of users before a phased rollout (Zijing Wu/Financial Times) ref-72 TechMeme GitHub Copilot's new pricing model went into effect today, and many noted sticker shock with some saying a few hours of AI usage ate big chunks of monthly caps (Kyle Orland/Ars Technica) ref-73 TechMeme Nvidia launches Nemotron 3 Ultra, a 550B-parameter MoE open model; Artificial Analysis: it's the smartest open US model but trails the Chinese model Kimi K2.6 (Maximilian Schreiner/The Decoder) ref-74 TechMeme Salesforce is acquiring CMS provider Contentful; source: Salesforce paid between $1B and $1.5B, a steep discount from Contentful's $3B valuation in 2021 (The Information) ref-75 TechMeme Startups from the pre-ChatGPT era face a reckoning in private markets; PitchBook: half of US unicorns haven't raised in three years, 220+ are "fallen unicorns" (Hugh Son/CNBC) ref-76 InfoQ AI/ML Google Workspace CLI: Unified Command-Line Tool Built for Humans and AI Agents ref-77 InfoQ AI/ML Java News Roundup: OpenJDK JEPs, Hazelcast, Quarkus, Hibernate, Koog, JHipster, Introducing Endive ref-78 InfoQ AI/ML Claude Code Adds Dynamic Workflows for Parallel Agent Coordination ref-79 InfoQ AI/ML Shopify Reports 15X Faster Graphql Execution with Breadth First Engine ref-80 InfoQ AI/ML BadHost Vulnerability Exposes AI Agents, Evaluators, and LLM Gateways ref-81 InfoQ AI/ML Presentation: Theme Systems at Scale: How To Build Highly Customizable Software ref-82 InfoQ AI/ML Article: The AI Productivity Paradox in Test Automation: Moving Beyond Structural Validation to Perception and Intent ref-83 InfoQ AI/ML Podcast: Requirements Analysis for Architects: A Conversation with Sonya Natanzon ref-84 The New Stack OpenAI, Anthropic, Google, Amazon, and xAI all fail on type of attack, study finds ref-85 The New Stack JetBrains open-sources Mellum2 to go where Claude Code can’t ref-86 The New Stack Claude Code vs. Cursor vs. Codex vs. Antigravity — six months in ref-87 The New Stack This coding agent doesn’t want your feedback — it ships without it ref-88 The New Stack “Blowing things up”: The one move vendors got wrong on AI agents ref-89 The New Stack At Sapphire, SAP makes the case that enterprise AI is a context problem ref-90 Wired AI Anthropic Confidentially Files for What Could Be the Largest IPO Ever ref-91 Wired AI Norse Atlantic Airways Offers Dirt-Cheap Tickets. There’s a Catch ref-92 TechCrunch Startups How to make the Startup Battlefield Top 20 — and what every company gets regardless ref-93 TechCrunch Startups Defense tech darling Mach Industries hits $1.8B valuation, a 4x jump in a year ref-94 TechCrunch Startups From the stage to the future: Where are Startup Battlefield’s alumni now? ref-95 TechCrunch Startups Revolut rolls out services to thousands of users in India ahead of broader launch ref-96 TechCrunch Startups Unastella, a South Korean rocket startup that launched from home, raises $24M ref-97 Crunchbase News Anthropic Files Confidentially For IPO ref-98 Crunchbase News How I Raised $14M For My Startup When I Stopped Pitching And Started Speaking ref-99 Crunchbase News Boston Startup Fundraising Looks Strong Only By Pre-AI Parameters ref-100 Sequoia Blog Listen to the Market ref-101 Tom's Hardware Watch Intel's Computex 2026 keynote here — CEO Lip-Bu Tan takes the stage in Taipei at 10:30pm PT on June 1 ref-102 Tom's Hardware Hands-on with Asus’ ROG Azoth Extreme Edition 20 mechanical keyboard – when carbon fiber isn’t enough, Asus goes for the gold ref-103 Tom's Hardware Hands-on with Asus’ ROG Harpe II Extreme Edition 20 gaming mouse – 24K gold and a 65K sensor ref-104 Tom's Hardware Asus shows off updated Zenbook and Strix Scar laptops, along with a TUF-based Gaming desktop — a refreshed look on laptops takes center stage ref-105 Tom's Hardware AMD Radeon RX 9070 GRE review: thoroughly midrange ref-106 Tom's Hardware Gigabyte debuts fourth-gen Tandem WOLED and multi-mode Mini LED gaming monitors — 27 to 32 inches, up to 480 Hz, and up to 5K resolution ref-107 Tom's Hardware Noctua announces new thermal pad for AMD chips in partnership with Carbice — product will work with processors in AM5 and AM4 sockets ref-108 Tom's Hardware Windows Server vulnerability can grant system privileges with just a malformed packet — domain controllers are being exploited in the wild ref-109 Tom's Hardware Anthropic files for IPO — Claude maker races OpenAI and SpaceX to Wall Street ref-110 Tom's Hardware Computex 2026 Day Zero Wrap-Up: Nvidia launches RTX Spark Superchip assault on laptop and desktop markets, Intel readies Xeon 6+ ref-111 Tom's Hardware Nvidia's RTX Spark could caplitalize where Qualcomm's Arm-based efforts have not — following the expiration of Qualcomm's Windows on Arm deal, Nvidia stands poised to pick up the slack ref-112 Tom's Hardware Qualcomm aims Snapdragon C laptop chip at the budget laptop segment, as manufacturers feel the DRAM squeeze — analysts warn sub $500 laptop market may disappear before 2028 ref-113 Tom's Hardware Access Tom’s Hardware Premium’s Computex 2026 coverage for free — sign up for an account to read insider reports from the show ref-114 Tom's Hardware Asus' world-first OLED esports monitor can hit 540Hz at 1080p — ROG Strix OLED model among four fresh offerings ref-115 Tom's Hardware Asus rolls out a ROG 20th anniversary chair and backpack, alongside commemorative components and peripherals — ROG Destrier Edition 20, ROG SLASH Hard-case Luggage Edition 20 are back in black (and gold) ref-116 Tom's Hardware Seven hospitalized after toxic gas fire at SK hynix advanced memory plant — Cheongju 4th campus incident today led to all 3,600 staff being evacuated ref-117 Tom's Hardware Bambu Lab Launches Big Bed Slinger: A2L — company's 'H2S Lite' is half the cost of H2S at just $469 ref-118 Tom's Hardware 256GB of dual-channel RAM hits mass market thanks to Origin Code — quad-rank CUDIMM packs 128GB of DDR5-8000 into a single module ref-119 Tom's Hardware Tom's Hardware Unfiltered: Computex 2026, Day 0 — peek behind the curtain to see how we're covering the biggest trade show of the year ref-120 Tom's Hardware Nvidia says RTX Spark chip will support all major anti-cheat and DRM technologies — Fortnite, Valorant, Denuvo, and more to work natively with Windows on Arm ref-121 Tom's Hardware Asus' monstrous ROG Astral GeForce RTX 5090 Edition 20 includes expansive curved AMOLED display — also debuts 3,000W power supply and striking PC case ref-122 Tom's Hardware Supermicro shows off Vera Rubin NVL72 rack with all-new type of coolant — company claims coolant offers 1,000 times higher electrical impedance over standard cooling ref-123 Tom's Hardware Save 78% on Nord's latest Complete VPN package — 27 months of online protection for $107 ref-124 Tom's Hardware US closes loophole that allowed Chinese-owned subsidiaries located outside China to buy AI chips — report claims that hundreds of thousands of advanced AI chips have been acquired through BIS blind spot ref-125 Tom's Hardware Negative time experiment clears peer review as photons appear to leave an atom cloud before entering — groundbreaking quantum 'negative time' proven after 1 million test runs ref-126 Tom's Hardware Acer’s PM131QT portable monitor is a 12.3-inch touchscreen with magnetic mounting, a built-in kickstand, and 5-point touch – 1920 x 720 IPS screen has pogo pins for a keyboard, and is designed for secondary and “in-vehicle” use ref-127 Tom's Hardware Resourceful runner 'can race my own ghost' using homemade Meta Ray-Ban Display app — also adds bonus coins, mini leaderboard, and more ref-128 Tom's Hardware Computex 2026 Live: Every update and announcement from day one in Taipei ref-129 Tom's Hardware AMD promises 13% uplift with new EXPO ‘Ultra Low Latency’ overclocking on DDR5 DIMMs — automatic memory overclocking delivers 4% improvement over standard EXPO, says AMD ref-130 Tom's Hardware Intel warns it has 'a healthy dose of paranoia' over Nvidia entrance into PC market — company says RTX Spark is 'great for the market' while touting the virtues of x86 ref-131 Krebs on Security Hackers Used Meta’s AI Support Bot to Seize Instagram Accounts ref-132 The Hacker News (Security) Dashlane Discloses Brute-Force Attack, Encrypted Vaults of Fewer Than 20 Users Downloaded ref-133 The Hacker News (Security) Miasma Supply Chain Attack Compromises Red Hat npm Packages with Credential-Stealing Worm ref-134 The Hacker News (Security) ⚡ Weekly Recap: New Linux Flaw, PAN-OS Exploit, AI-Powered Attacks, OAuth Phishing and More ref-135 The Hacker News (Security) China-Aligned Groups Ramp Up Attacks: Dragon Weave Hits Czech Republic & Taiwan ref-136 The Hacker News (Security) The Security Growth Platform: Why MSPs Are Moving Beyond vCISO Tools ref-137 The Hacker News (Security) OpenAI Codex Authentication Tokens Stolen in codexui-android npm Supply Chain Attack ref-138 EU Startups SEALSQ acquires majority stake in Wecan Group and invests €5.4 million to scale post-quantum financial security solutions ref-139 EU Startups Munich’s Bayshore exits stealth with €6.9 million to automate legal and compliance workflows with AI ref-140 EU Startups UK-based Circular11 secures €2.7 million to turn low-grade plastic waste into building materials ref-141 EU Startups Ask these three questions before choosing a co-founder or regret it later ref-142 MIT Research New propulsion system could make tiny satellites both fast and fuel-efficient ref-143 MIT Research Enzymes that assemble into droplets can speed up cellular reactions ref-144 Lobsters thunderbolt-ibverbs: We have InfiniBand at home ref-145 Lobsters Constraining LLMs Just Like Users ref-146 Google News Microsoft and Google are late to AI coding, but 'absolutely critical' they compete for growth ref-147 Google News This AI model backdoor attack stays hidden until you customize the model ref-148 Google News An AI model decodes tumor mutations to predict treatment response ref-149 Google News Nvidia chases $200B CPU market with AI agent PCs from Microsoft, Dell, and HP ref-150 Google News Starbucks quietly retired its AI agent just months after deployment after it miscounted coffee shop inventories and slowed down baristas ref-151 Google News FinancialContent - Salt Security launches Salt Code, the first agentic security solution to enforce security policies inside AI coding assistants ref-152 Google News Tencent moves closer to launching AI agent for China’s most-used app ref-153 Google News Anthropic is giving the E.U.'s cybersecurity agency access to its restricted AI security model ref-154 Google News Turing Award winner Richard Sutton says pure generative AI can't do real science ref-155 Google News AI Agent Economics: Token Tax Locks Gross Margins 30 Points Below SaaS Baseline ref-156 Google News AI model finally learns to say ‘I don’t know’ in breakthrough to curb chatbot overconfidence ref-157 Google News Singapore rolls out AI agent registry as govt expands use of artificial intelligence in public service ref-158 Google News Meituan (03690.HK): AI agent "Xiaomei" to launch in collaboration with Tencent Yuanbao ref-159 Google News A Chinese-made AI model, 'MiniMax M3,' has appeared, a high-performance open-source model that competes with GPT-5.5 and Claude Opus 4.7. ref-160 Google News Amazon's AI jackpot: Its $8 billion Anthropic holdings are now worth $74 billion ref-161 Google News Anthropic files for IPO as AI startup behind ‘Claude’ prepares historic US stock market debut ref-162 Google News Workday, Google Cloud expand partnership to bring AI agents into HR and finance workflows ref-163 Google News CNN Sues Perplexity Over 17,000 Scraped Assets; the AI Startup Now Faces Six Active Publisher Lawsuits ref-164 Google News India-US trade talks underway as semiconductor and AI partnership gains pace ref-165 Google News Anthropic Hits $965B Valuation, Overtakes OpenAI as Most Valuable AI Startup ref-166 Google News Our views on AI policy and political advocacy ref-167 Google News Florida is suing OpenAI and Sam Altman, making it the first state to target the company over AI safety ref-168 Google News Redefining AI Safety for a Multimodal and Multicultural World ref-169 Google News Pope Leo XIV Issues Encyclical Calling for AI Regulation ref-170 Google News South Africa Sets January 2027 Public Comment Date for Revised National AI Policy ref-171 Google News Arm Holdings may hit $15 billion AI chip revenue goal early, CEO says - Bloomberg News ref-172 Google News Nvidia RTX Spark comes to Windows PCs with Arm CPU, RTX GPU, and unified memory ref-173 Google News Intel unveils Xeon 6+ to power agentic AI inference, challenges GPU-centric infrastructure ref-174 Google News Alphabet announces $80 billion equity raise to fund AI infrastructure, including a $10 billion bet from Berkshire Hathaway ref-175 Google News Nvidia Vera Rubin AI chip platform enters full production