Reports
Report 2026-07-09 167 sources Input Tokens: 109,621 · Output Tokens: 3,798 · Total Tokens: 115,729

AI 戰情報告 (2026-07-09)

openaianthropicxaimetanvidiask-hynixcognitionrefiant grok-4.5gpt-5.6-solclaude-4.8swe-1.7proteamuse-image agentcontext-windowcybersecuritysemiconductordevopsllm-benchmarking

AI 戰情報告 (2026-07-09)

昨日 AI 重點新聞

xAI 正式發布 Grok 4.5:定位編碼與代理的「Opus 殺手」 ref-2, ref-38, ref-6

Elon Musk 宣布 Grok 4.5 正式向大眾開放。該模型由 1.5 兆參數 的 V9 基礎模型驅動,專為 AI 代理 (Agent) 與編碼任務優化,並深度整合 Cursor 編碼器。Musk 稱其為「Opus 等級」,但在速度與 Token 效率上更具優勢,定價為每百萬輸入 Token 2 美元。

OpenAI GPT-5.6 Sol 系列全球上線並推出 GPT-Live 語音模型 ref-106, ref-148, ref-14, ref-16

OpenAI 結束為期數週的政府審查,正式推出次世代旗艦模型 GPT-5.6 Sol,以及更具性價比的 Terra 與 Luna。同步推出的 GPT-Live 採用全雙工架構,支援即時語音對話並能背景調用 Sol 執行深層推理,展現強大的語音助理能力。

Meta 推出 Muse Image 影像模型並傳出 Watermelon 性能追平 GPT-5.5 ref-6, ref-8, ref-108

Meta 發表全新的 Muse Image 模型,其最大特點是預設允許使用 Instagram 的公開照片進行 AI 影像生成。此外,內部傳出代號 Watermelon 的模型在基準測試中已趕上 GPT-5.5,顯示 Meta 正全面追趕閉源領先者。

Cognition 發布 SWE-1.7:每秒 1,000 Token 的工程暴力美學 ref-6, ref-57

Cognition 發表 SWE-1.7,該模型基於 Kimi K2.7 訓練,在維持高智慧水平(逼近 GPT-5.5)的同時,能以驚人的每秒 1,000 Token 速度產出。其目標是為企業提供極速且低成本的自動化開發解決方案。

Refiant 發表 Protea:實現 1,000 萬 Token 超長上下文窗口 ref-76, ref-139

新創公司 Refiant 推出 Protea 模型,運用「群體優化 (Swarm Optimization)」技術將上下文長度推升至 1,000 萬 Token。這相當於一次讀入 83 本小說或 830 集播客,旨在讓模型直接在工作記憶中處理整份企業代碼庫,減少對 RAG 的依賴。

資安警訊:Friendly Fire 與 GhostApproval 漏洞威脅 AI 代理安全 ref-110, ref-8, ref-109

資安界揭露兩大針對編碼代理的嚴重漏洞:GhostApproval 透過符號連結 (Symlink) 讓代理在不知情下修改敏感檔案;Friendly Fire 則能誘導代理運行攻擊者的惡意代碼。受影響工具包含 Claude Code、Amazon Q 與 Cursor。

Anthropic 為 AWS 推導 Claude 應用閘道器 (Gateway) ref-8

Anthropic 與 AWS 合作推出 Claude Apps Gateway,為企業提供單一控制平面,用以管理大規模部署的 Claude Code 與 Claude Desktop。此工具支援中心化的成本控管、安全策略執行與身份驗證。

前 GitHub 執行長創立 Entire:建構代理時代的分布式 Git 網路 ref-77, ref-141

Thomas Dohmke 創立 Entire,目標是解決大量 AI 代理頻繁 Push/Pull 對中央伺服器造成的負擔。Entire 重新實作了 Git 後端,提供比 GitHub 快 25 倍 的推送速度,並支持全球分布的節點存取。

SK Hynix 280 億美元增資案獲 7 倍超額認購 ref-165

受惠於 AI 晶片熱潮,韓國半導體巨頭 SK Hynix 的 280 億美元股權出售案吸引了市場高度關注,需求達供應量的 7 倍,顯示全球資本對 HBM 記憶體供應鏈的極度渴求。

QuantumDiamonds 獲 9,100 萬歐元:晶片檢測界的下一個 ASML ref-82, ref-122

德國量子檢測新創 QuantumDiamonds 完成大額融資,利用量子感測技術在奈米尺度下偵測晶片缺陷。該公司已在台灣新竹設立亞太中心,目標是為台積電等頂尖晶圓廠提供次世代檢測設備。

努比亞 (Nubia) 發布全球首款「AI 代理手機」 ref-152

手機品牌 Nubia 搶先於 WAIC 前夕發表號稱全球首款內建 AI Agent 的智慧型手機,強調 AI 能自主執行跨 App 任務,標誌著 On-device AI 邁向代理化營運階段。

NVIDIA Nemotron 結合 LangChain 達成商業任務效能躍升 ref-22

NVIDIA 宣布其 Nemotron 3 Ultra 透過 LangChain Deep Agents 框架優化,能在不調整模型參數下,僅靠環境工程 (Harness Engineering) 就達到與頂尖閉源模型相當的準確度,且推理成本降低 10 倍。

AI 輔助軟體重寫新範式:Bun 採用 Rust 重寫成功 ref-33

Bun 創辦人分享利用 AI 代理(耗資 16.5 萬美元 Token 費)在 11 天內將核心由 Zig 遷移至 Rust 的經驗,證明了在強大測試套件與代理協作下,大型代碼遷移的經濟效益已發生質變。

市場洞察

1. 「代理基礎設施 (Agent Infra)」進入標準化與治理階段

延續昨日報告中對 Vercel 收購 Better Auth 的觀察,今日 Entire 的分布式 Git (ref-50)、Anthropic 的 AWS 閘道器 (ref-29) 以及 NVIDIA 的 Nemotron Harness (ref-44),共同揭示了產業正快速從「開發代理」轉向「治理代理」。企業已不滿足於單純的代理工具,而是需要一套能應付「代理洪流(Agent Swarms)」的高效、安全且可歸責的底層架構。

2. 「超長上下文」與「高效 RAG」的典範競爭

Refiant Protea 實現的 1,000 萬 Token 窗口 (ref-43, ref-76) 挑戰了傳統 RAG 架構。雖然 RAG 具備低延遲優勢,但 Protea 的群體優化技術顯示,直接在模型內部進行全量資料推理正變得可行。這意味著未來半年內,企業在處理「百萬行代碼庫」或「十年財報資料」時,將面臨選取長上下文模型或精細 RAG 系統的技術路徑決策。

3. 效能與成本之戰:向「百萬分之一價格」挺進

今日 OpenAI Sol/Terra/Luna (ref-81) 與 Grok 4.5 (ref-26) 的同步發布,不僅是性能的比拼,更是極致的價格戰。隨著 Terra 提供比 GPT-5.5 便宜一半的價格,以及 SWE-1.7 提供極速推理 (ref-11),市場正向「高智慧、低延遲、亞美分計價」快速演進。這種趨勢將迫使所有垂直領域的 AI 應用必須重新評估其利潤結構。

4. 硬體供給側的強韌需求與量子技術介入

SK Hynix 的融資盛況 (ref-0) 與 QuantumDiamonds 的崛起 (ref-21) 顯示,AI 晶片的競爭已深入到製造端的良率檢測與材料科學。隨著先進封裝與 3D 晶片架構日益複雜,量子等級的檢測設備將成為台積電等晶圓廠能否維持產能與利潤的關鍵「軍火商」。

Sources 167 items
ref-0 Hacker News John Deere owners will get the right to repair equipment under FTC settlement ref-1 Hacker News GPT‑Live ref-2 Hacker News Grok 4.5 ref-3 Hacker News Mistral's Robostral Navigate: a state of the art robotics navigation model ref-4 Hacker News I think I have LLM burnout ref-5 Hacker News Show HN: Microsoft releases Flint, a visualization language for AI agents ref-6 Hacker News SWE-1.7 Reach Near GPT 5.5 and Opus Intelligence ref-7 Hacker News Show HN: Yamanote.fun – A complete soundscape for Tokyo's Yamanote line ref-8 Hacker News We made Grok 4.5, GPT-5.5, and Claude build the same apps ref-9 Hacker News Benchmarking coding agents on Databricks' multi-million line codebase ref-10 Hacker News MIRA: Multiplayer Interactive World Models Trained on Rocket League ref-11 Hacker News AI changes the economics of software rewrites ref-12 Hacker News Show HN: Frugon – Find which LLM calls a cheaper model could handle (local, MIT) ref-13 Hacker News Ergo: Long Form Philosophy Lectures ref-14 OpenAI Blog Our approach to government and national security partnerships ref-15 OpenAI Blog Separating signal from noise in coding evaluations ref-16 OpenAI Blog Helping K–12 educators build practical AI skills ref-17 Google Developers Blog Bridging the Domain Gap: AI Race Coach built with Antigravity and Gemini ref-18 Google Developers Blog Driving the Agent Quality Flywheel from Your Coding Agent ref-19 Google Developers Blog Build reliable multi-agent applications with ADK Go 2.0. Discover our new graph-based workflow engine, built-in human-in-the-loop, and dynamic orchestration ref-20 Google Developers Blog Measuring What Matters with Jules ref-21 Google Developers Blog Build Cross-Language Multi-Agent Team with Google’s Agent Development Kit and A2A ref-22 Hugging Face Blog Data for Agents ref-23 NVIDIA AI Blog NVIDIA Nemotron Achieves Benchmark-Leading Performance With LangChain Deep Agents Harness ref-24 AWS ML Blog Introducing Claude apps gateway for AWS ref-25 AWS ML Blog Powering scientific discovery: BYOKG and GraphRAG for intelligent pharmaceutical research ref-26 AWS ML Blog Automatically sort and prioritize your mailboxes by using Amazon Bedrock ref-27 AWS ML Blog Building and connecting a production-ready ecommerce MCP server using Amazon Bedrock AgentCore and Mistral AI Studio ref-28 AWS ML Blog Securing Amazon Bedrock AgentCore Runtime with AWS WAF ref-29 AWS ML Blog Manage AI applications on Mac with Jamf’s AI Governance and Amazon Bedrock ref-30 The Rundown AI Meta climbs the AI image leaderboard ref-31 Latent Space [AINews] SpaceXAI launches Grok 4.5, first Opus-class model post Cursor acquisition ref-32 Latent Space Why AI Infrastructure must evolve for Agent Experience — Akshat Bubna, Modal CTO ref-33 Simon Willison Rewriting Bun in Rust ref-34 Simon Willison Introducing GPT‑Live ref-35 Simon Willison Quoting Kenton Varda ref-36 TechCrunch AI Lovable reportedly in talks to double its valuation to $13.2B ref-37 TechCrunch AI Google’s deepfake detector system used to debunk McConnell hoax pic ref-38 TechCrunch AI SpaceXAI releases Grok 4.5, which Elon describes as an ‘Opus-class model’ ref-39 TechCrunch AI This startup thinks robotics is about to have its ChatGPT moment ref-40 TechCrunch AI Google Photos adds a new AI ‘Video Remix’ tool ref-41 TechCrunch AI Why this CEO thinks video games make better training data than the internet ref-42 TechCrunch AI Meta wants its AI glasses to seem less creepy. Its AI strategy says otherwise. ref-43 TechCrunch AI OpenAI releases new voice models for more natural live conversations ref-44 TechCrunch AI Prime Intellect raises $130M Series A to help enterprises build their own AI agents ref-45 TechCrunch AI These AI startups are growing revenue at faster and faster rates ref-46 TechCrunch AI Your gaming data could be the secret to AGI, according to this Bezos-backed startup ref-47 TechCrunch AI Former OpenAI exec Kevin Weil is now on the board of Stoke Space ref-48 The Verge AI ChatGPT’s upgraded voice mode is better at shutting up ref-49 Ars Technica Google pays $250K for Linux vulnerability allowing guest VM escapes ref-50 Ars Technica Aussie gov't tells volunteers to throw out thousands of functioning test routers ref-51 Ars Technica US rare earths flow to Asia as domestic demand is slow to emerge ref-52 TechMeme The AI data center boom has led to surging demand for power transformers, with average lead times for orders, once measured in months, now stretching into years (Financial Times) ref-53 TechMeme A Brown University professor suspected his class used AI to cheat after a take-home midterm averaged 96%, prompting an in-person final, which averaged 48.6% (Emma Whitford/Inside Higher Ed) ref-54 TechMeme Companies are mobilizing internal groups of "AI champions" to drive adoption; BCG says 74% of front-line employees now use AI regularly, up from 51% in 2025 (Isabelle Bousquette/Wall Street Journal) ref-55 TechMeme Shanghai-based GPU maker Iluvatar CoreX raised ~$902M in a Hong Kong share sale; the company's stock has soared 257% since its January IPO, which raised $473M (Bloomberg) ref-56 TechMeme Filings: Grubhub parent Wonder is raising hundreds of millions at a $9B valuation; source: founder Marc Lore indicated he'll contribute $200M to the round (The Information) ref-57 TechMeme Cognition releases SWE-1.7, trained from Kimi K2.7 and available at 1,000 tokens/second, claiming it nears GPT-5.5 and Opus 4.8 on benchmarks at a lower cost (Cognition) ref-58 TechMeme Block agrees to pay $45M and offer live customer support for Cash App to settle claims by 46 US states that the company failed to protect users from fraud (Daniel Wiessner/Reuters) ref-59 TechMeme Sources: Reno-based AI chip startup Positron is in talks to raise ~$750M in two phases, at valuations of $3.5B in the first tranche and ~$5B in the second (Bloomberg) ref-60 TechMeme The FTC settles with John Deere over a 2025 right-to-repair lawsuit, requiring the company to provide farmers and shops with equipment and software for 10 years (Boone Ashworth/Wired) ref-61 TechMeme OpenAI says it found widespread task issues in SWE-Bench Pro, estimates ~30% of tasks are broken, and retracts its earlier recommendation to adopt the benchmark (OpenAI) ref-62 TechMeme Meta announces its first data center in Canada, a 1GW facility in Alberta that will cost the company about $9B and take two to three years to construct (Jonathan Vanian/CNBC) ref-63 TechMeme Grok 4.5 is available in Grok Build, in Cursor, and from the SpaceXAI console, but not in the EU; it costs $2 per 1M input tokens and $6 per 1M output tokens (Axios) ref-64 TechMeme SpaceXAI launches Grok 4.5, its first model built in partnership with Cursor, designed to "handle difficult, long-running" legal, finance, and coding tasks (Carmen Arroyo/Bloomberg) ref-65 TechMeme OpenAI rolls out two versions of GPT-Live: GPT-Live-1, powering ChatGPT Voice for Go, Plus, and Pro users, and GPT-Live-1 mini, the default for free users (Sabrina Ortiz/The Deep View) ref-66 TechMeme OpenAI launches GPT-Live, a new generation of voice models built on a full-duplex architecture, meaning they can listen and speak at the same time (OpenAI) ref-67 InfoQ AI/ML AWS Details How One Customer Scaled to One Million Lambda Functions ref-68 InfoQ AI/ML The Kubernetes Approach to AI-Assisted Maintainership Prioritises Human Accountability ref-69 InfoQ AI/ML Airbnb Shares Architecture Behind Sitar-Agent Dynamic Configuration Sidecar for Kubernetes Services ref-70 InfoQ AI/ML Presentation: The Multi-Agent Approach: Building Reliable and Controllable Software Development Automation ref-71 The New Stack “Opus-class, but faster”: What Elon Musk says about beating Anthropic ref-72 The New Stack JetBrains’ next move isn’t a better IDE — it’s a governance layer over Claude Code, Codex, and Gemini CLI ref-73 The New Stack Meta says it caught OpenAI. One thing is missing. ref-74 The New Stack OpenAI’s own safety card says GPT-5.6 has a lying problem ref-75 The New Stack Most enterprises will hand root cause analysis to AI agents within two years ref-76 The New Stack “Nature is the most computationally efficient system we know”: How Refiant used swarm optimization to build a 10-million-token AI model ref-77 The New Stack Entire is building a Git network for agents ref-78 Wired AI Messi and Ronaldo Are Building Tech Portfolios. Mo Salah Is Playing a Different Game ref-79 Wired AI I Built a Self-Improving AI, and So Can You ref-80 Wired AI Pickup Artist Mystery Has an AI Girlfriend ref-81 Wired AI This Former DeepMind Exec Thinks the AI Arms Race Could End in Disaster ref-82 TechCrunch Startups With EU backing, QuantumDiamonds aims to speed up chip manufacturing ref-83 Crunchbase News Exclusive: EdVisorly Raises $13.3M Series A To Fix The Messy College Transfer Process With AI ref-84 Crunchbase News Familiar Names Top Active US Investor Ranks In Q2 ref-85 Crunchbase News Your SaaS Metrics Are A Result, Not A Strategy ref-86 Stratechery XBOX Cuts; Bundling and the Internet Solvent; Transaction, Coordination, and Sunk Costs ref-87 Tom's Hardware Rapidus fab roadmap examined — first new leading-edge chipmaker in decades has one Hokkaido fab, a 2027 deadline, and 60 potential customers ref-88 Tom's Hardware China alleges that Claude Code contains backdoors, calls mechanism 'a serious threat' — Gov't claims Claude sends sensitive information to remote servers without consent ref-89 Tom's Hardware Hidden backdoor in Tenda routers goes unpatched as company ignores warnings from cybersecurity researchers — Chinese company's firmware allows admin access without a password ref-90 Tom's Hardware Budget smartphone market collapses under the weight of memory shortages, sales expected to drop 22% — memory alone now comprises up to 64% of the total cost of lower-tier smartphones ref-91 Tom's Hardware JEDEC releases new SPHBM4 standard to slash AI memory costs — Narrow 512-bit interface enables dropping expensive interposers for organic substrates ref-92 Tom's Hardware Global consumer Wi-Fi router shipments fell 6% in Q1 2026, down 34% from 2021 peak — mesh systems and gaming routers still prove popular ref-93 Tom's Hardware SiPearl's long-awaited Rhea CPU finally gets in the lab, opening the door for Europe's first sovereign HPC CPU — 'availability of Rhea1 is scheduled for end of 2026' SiPearl VP says, following long development process ref-94 Tom's Hardware Power company hikes data center bills by 30%, cuts residential electricity costs by 1.3% — Oregon approves change through POWER Act, pushes developments using more than 20 Megawatts of power to pay their fair share ref-95 Tom's Hardware 'Slopfix' software team charges $10,000 a week to delete AI-generated code bloat — ironically, the team uses AI agents to trim messy repositories by up to 65% ref-96 Tom's Hardware Fi Ultra becomes first dog tracker powered by Starlink satellites – the Fi Ultra Dog Tracker makes Fido trackable via satellite, onboard GPS, and LTE connectivity ref-97 Tom's Hardware Cooler Master V4 and V8 3DHP Review: A masterful engineering achievement ref-98 Tom's Hardware Cracked version of Assassin’s Creed Black Flag Resynced leaked days prior to official release despite Denuvo DRM protection — Denuvo unable to stop crackers, with some finding ways to completely remove it from other titles ref-99 Tom's Hardware Modding tool 'DLSS Swapper' might infect your PC with malware if you download the wrong files — App creator warns against using random, user-submitted DLLs ref-100 Tom's Hardware Save $527 on this 1440p gaming PC from Newegg with a 16GB RTX 5060 Ti right now — huge discount for high-end performance rig with an Intel Core i7-14700F, 32GB DDR5, and a 1TB SSD ref-101 Tom's Hardware Valve releases drivers, notes to make Windows work on Steam hardware, but refuses to support it — tells users it doesn’t offer support for ‘Windows on Steam Hardware,’ gaming company provides resources ‘as is’ ref-102 Tom's Hardware 8Bitdo's wireless Pro 2 gaming controller falls to all-time low price — hall-effect gamepad is 38% off, just $37.19 ref-103 Tom's Hardware Nvidia touts Vera CPU's single-threaded performance as its agentic AI advantage, reveals next-gen 'Rigel' Arm CPU cores — frames chip as a 'max single-threaded CPU at scale,' not a parallel monster ref-104 Tom's Hardware Arrest and extradition of Scattered Spider hacker shines light on how Windows telemetry GDIDs can identify and track users — Microsoft device identifier is just one digital fingerprint in a software world rife with them ref-105 Tom's Hardware Space Force gets first mobile high-powered electromagnetic beam weapon to cripple enemy satellites — plans to deploy 32 ‘Meadowlands’ units to detect, deny, disrupt, and degrade hostile space assets ref-106 Politico Tech OpenAI to release its most powerful model after weekslong hold ref-107 Krebs on Security Felons, Fraudsters Flog Offensive Cybersecurity Startup ref-108 The Hacker News (Security) Meta's New AI Image Tool Lets Others Use Your Public Instagram Photos in AI Images ref-109 The Hacker News (Security) Top AI Agents Built to Catch Malicious Code Can Be Tricked Into Running It ref-110 The Hacker News (Security) GhostApproval Symlink Flaws Could Let Malicious Repos Run Code in AI Coding Agents ref-111 The Hacker News (Security) Fake 7-Zip Installers Turn Devices Into Residential Proxy Nodes ref-112 The Hacker News (Security) AI Coding Agents Found Triggering Endpoint Security Rules Built to Catch Attackers ref-113 The Hacker News (Security) New HalluSquatting Attack Could Trick AI Coding Assistants Into Installing Botnet Malware ref-114 The Hacker News (Security) Ubiquiti Patches Critical UniFi Flaws Across Connect, Talk, Access, Protect, and OS ref-115 The Hacker News (Security) New Ghost Phishing Wave Is Breaking Traditional Email Security ref-116 The Hacker News (Security) SCMBANKER Malware Uses ClickFix Lures to Target Mexican Banking Users ref-117 The Hacker News (Security) GitHub 'Verified' Commits Can Be Rewritten Into New Hashes Without Breaking Signatures ref-118 The Hacker News (Security) The Verification Step Is the New ATO Battleground in 2026 ref-119 The Hacker News (Security) GitHub Copilot Refuses Harmful Requests in Chat, Then Writes Them in Code ref-120 The Hacker News (Security) China-Linked UAT-7810 Expands ORB Network With New LONGLEASH Malware ref-121 EU Startups Alchemab Therapeutics secures €29.3 million as British Business Bank makes its largest life sciences investment yet ref-122 EU Startups Munich’s QuantumDiamonds raises €91 million to scale its quantum-based semiconductor inspection technology ref-123 EU Startups Can entrepreneurs have children and still succeed? New survey digs deep ref-124 EU Startups Why AI provider dependency is becoming Europe’s next startup risk (Sponsored) ref-125 EU Startups Toulouse-based Bohr Energie secures €10 million to scale its AI-powered renewable energy aggregation platform ref-126 EU Startups As power demand rises from electrification and data centres, London’s Axle Energy raises €21 million for energy flexibility ref-127 EU Startups MediaFuse launches TechnologyWire, a press release distribution platform built for the Tech industry ref-128 EU Startups How EIC-backed startups are turning international exposure into real business ref-129 EU Startups YC-backed Fleek raises €22 million to build AI infrastructure keeping global secondhand fashion on fleek ref-130 MIT Research Separating logic and language ref-131 MIT Research MIT engineers whip up a more breathable hydrogel ref-132 MIT Research MIT researcher proposes a way to detect nuclear weapons in space ref-133 Lobsters Native-speed vLLM transformers modeling backend ref-134 Google News When Your Factory Worker Is An AI Model: What Boards Need To Know About Physical AI ref-135 Google News GhostApproval: A Trust Boundary Gap in AI Coding Assistants ref-136 Google News Tencent Tests AI Agent for WeChat's 1.4 Billion Users ref-137 Google News GitHub AI agent leaks private repositories via prompt injection attack ref-138 Google News SpaceXAI Launches Grok 4.5—Here’s What’s New About The Company’s ‘Strongest Model Ever’ ref-139 Google News Refiant goes where rivals only promised with a 10 million-token AI model ref-140 Google News U.S., China Restrict AI Model Exports ref-141 Google News Ex-GitHub chief’s Entire opens distributed Git network for the AI agent era ref-142 Google News Anthropic expands Claude Cowork to mobile and web, letting AI agent work across devices ref-143 Google News China warns of "security backdoor" in Anthropic AI coding tool ref-144 Google News Trump's EO Puts AI Agent Governance On GC Agenda ref-145 Google News Agentic Botnets Attack Uses HalluSquatting to Hijack AI Coding Assistants ref-146 Google News New AI model reveals how neutron star mergers forge heavy elements ref-147 Google News Perplexity picks Nvidia's new chip for AI agent tasks ref-148 Google News OpenAI to publicly release GPT-5.6 AI models, ending government-requested limits ref-149 Google News NVIDIA deploys AI agent for factory alarm triage ref-150 Google News OpenAI’s GPT-Live-1 challenges Google in AI model race ref-151 Google News AI agent carries out full ransomware attack for first time, researchers say ref-152 Google News World First AI Agent Phone Debuts: Nubia Claims Global First Ahead of WAIC Shanghai ref-153 Google News Microsoft hopes to cut its genAI costs by focusing on its own models ref-154 Google News This AI Startup Founder Just Pleaded Guilty in a Massive, 30-Person Insider Trading Scheme ref-155 Google News Legal AI startup Norm Ai hits $1.2 billion valuation after $120 million funding ref-156 Google News Chinese AI startup Zhipu seeks $4b share sale ref-157 Google News UK AI startup Nscale secures $900m credit facility ref-158 Google News AVGO Stock Alert: Broadcom Explodes On "Insane" $10.8B AI Revenue Beat! St Johns (84v84tv1Ts) ref-159 Google News Inaugural UN Global Dialogue on AI Governance ends with call to turn principles into action before 2027 ref-160 Google News Illinois Gov. JB Pritzker signs the country’s strongest AI regulation bill ref-161 Google News Meta paused AI training on employee keystrokes after data was 'put in a place it wasn't supposed to go:' CTO ref-162 Google News EXCLUSIVE: China's DeepSeek developing its own AI chip, sources say ref-163 Google News AI chip startup SambaNova valued at $11 billion in $1 billion funding round ref-164 Google News Amazon launches $25B bond sale to fund AI infrastructure ref-165 Google News SK Hynix’s $28 billion share sale gets 7x demand as AI chip frenzy refuses to cool ref-166 Google News China Eases AI Chip Restrictions for NVIDIA (NVDA), Unlocks H200 Supply Potential