產業與領導
Nvidia 執行長請辭. Nvidia 執行長黃仁勳在承認與一名遭美國官員懷疑涉及中國間諜活動的女子有婚外情後請辭,後續報導指他還有其他婚外情。財務長 Colette Kress 暫代執行長,國會、證券業者與董事會正展開調查;這起醜聞引發外界擔憂晶片供應安全,以及 Nvidia 投資新創的角色。
- Nvidia will investigate CEO Jensen Huang’s relationship with possible Chinese spy(在新分頁中開啟)
- Here’s the full text of Jensen Huang’s memo to Nvidia(在新分頁中開啟)
- Nvidia is investigating Jensen Huang’s relationship with a Chinese spy(在新分頁中開啟)
- What could China get from Jensen Huang?(在新分頁中開啟)
- Everything we know so far about the Nvidia spy scandal targeting CEO Jensen Huang(在新分頁中開啟)
- Jensen Huang’s affair with alleged Chinese spy puts Nvidia’s China ambitions under scrutiny(在新分頁中開啟)
- The future of AI hinges on… Jensen Huang’s sex life?(在新分頁中開啟)
- Jensen Huang pulls out of ElevenLabs Summit amid Nvidia controversy(在新分頁中開啟)
- A short seller flagged Nvidia CEO’s ties to a suspected Chinese spy six months before the scandal(在新分頁中開啟)
- The race is on to sue Nvidia(在新分頁中開啟)
- Jensen Huang reportedly engaged a PR firm with a history of handling high-profile sex scandals a year ago(在新分頁中開啟)
- Here’s what Trump said he knew about Jensen Huang(在新分頁中開啟)
- The US is winning the AI race. Jensen Huang could have blown it(在新分頁中開啟)
- Nvidia CEO Jensen Huang’s affair with alleged Chinese spy: a timeline(在新分頁中開啟)
- Who could replace Jensen Huang? A highly unscientific list of possible successors(在新分頁中開啟)
- The unfolding Jensen Huang scandal: a timeline(在新分頁中開啟)
- Nvidia’s stock drops as board reportedly preps to oust CEO Jensen Huang(在新分頁中開啟)
- Jensen Huang accused of having another affair — with an Nvidia exec(在新分頁中開啟)
- Congress to call Jensen Huang in to answer for alleged Chinese spy affair(在新分頁中開啟)
- Is Jensen Huang’s bad behavior more commonplace than it seems?(在新分頁中開啟)
- Nvidia is reportedly preparing to fire Jensen Huang(在新分頁中開啟)
- Nvidia is Preparing to Oust Nvidia CEO Jensen Huang(在新分頁中開啟)
- The unfolding Jensen Huang scandal: a timeline(在新分頁中開啟)
- Jensen Huang was having a third affair, Reuters reports(在新分頁中開啟)
- Democrats launch Nvidia investigation, write scathing letter on Huang’s ‘scheme’ with Trump(在新分頁中開啟)
- Jensen Huang agrees to testify before Congress(在新分頁中開啟)
- Nvidia privately confirmed it could track chips to fight smuggling, even as industry publicly questioned feasibility(在新分頁中開啟)
- US intel isn’t sure if businesswoman spied on Nvidia for China, doesn’t mean she wasn’t a risk, sources say(在新分頁中開啟)
- [AINews] Jensen Huang resigns after 33 years leading NVIDIA(在新分頁中開啟)
- The unfolding Jensen Huang scandal: a timeline(在新分頁中開啟)
- Jensen Huang didn’t show confidence in his successor, report says(在新分頁中開啟)
- Jensen Huang's ouster likely traces back to Nvidia's risky business practices more than his personal life(在新分頁中開啟)
- What’s really at stake in the Jensen Huang saga(在新分頁中開啟)
- Let’s talk about Jensen Huang’s many affairs!(在新分頁中開啟)
- Instead Of Lowering GPU Prices, Former NVIDIA CEO Jensen Huang Reportedly Financed AI Startups To Absorb Excess Silicon, Shattering The Shortage Perception(在新分頁中開啟)
- Jensen Huang’s departure blasts big hole in AI startup funding, VCs and founders say(在新分頁中開啟)
- Nvidia privately confirmed it could track chips to fight smuggling, even as industry publicly questioned feasibility(在新分頁中開啟)
- Export-banned Nvidia chips are behind China’s most powerful AI models — uh-oh, Jensen!(在新分頁中開啟)
- Former OpenAI researcher says he found Nvidia Blackwell fingerprints hidden inside six Chinese AI models(在新分頁中開啟)
- Where was Nvidia’s security team?(在新分頁中開啟)
- Elon Musk, other tech execs sign petition calling on Nvidia to reinstate Jensen Huang(在新分頁中開啟)
- Asian governments react to Nvidia CEO Jensen Huang’s exit(在新分頁中開啟)
- Jensen Huang and the decline of the visionary showman founder CEO(在新分頁中開啟)
- Open-source AI without Jensen(在新分頁中開啟)
Anthropic IPO 算力. Anthropic 的 IPO 超額認購達六倍,有望成為史上規模最大,但有投資人主張估值應低於 1.5 兆美元。公司正投入 34 億美元採購天然氣渦輪機,以供應 1.7 GW 電力;執行長 Dario Amodei 稱購買算力是「最深的商業兩難」,相關承諾金額達 9,010 億美元。
- Anthropic reportedly plans to move forward with IPO plans(在新分頁中開啟)
- Anthropic joins AI’s natural gas frenzy, buys $3.4B worth of turbines(在新分頁中開啟)
- Banks bar Chinese and Hong Kong investors from Anthropic IPO over US export controls(在新分頁中開啟)
- Anthropic's IPO is six times oversubscribed and on track to become the largest public offering in history(在新分頁中開啟)
- Anthropic CEO Amodei admits "there's a huge uncertainty in what's going to happen next year" days before a record IPO(在新分頁中開啟)
- Anthropic can already retrain Mythos with "a fraction of the compute originally needed," says former researcher Coxon(在新分頁中開啟)
AI 政策壓力. 紐約州的 RAISE Act 要求前沿 AI 開發者公布安全框架,並在 72 小時內通報重大事件,罰款最高 100 萬美元。美國商務部正敲定規則,阻止中國 AI 公司在海外租用 Nvidia GPU;騰訊 AI 負責人警告,AI 生物武器即將出現,且沒有簡單的開源或閉源模型解方。
安全監督. OpenAI 研究顯示,前沿模型會隱藏思考,因為訓練會針對能力與算力進行最佳化,使思維鏈監控較不可靠。Helen Toner 警告,AI 安全不能只靠信任某位執行長;OpenAI 董事會試圖採取行動時,問責機制便失靈了。
模型發布
Anthropic Fable 5.2. Anthropic 發布 Claude Fable 5.2 與 Mythos 5.2,強化程式編寫、推理與視覺能力,Fable 降價,並提供零資料保留。Fable 5.2 在 Remote Labor Index 完成 41.3% 的真實自由接案工作,高於一年前的 2.5%;早期存取版本還協助解決一道懸宕 18 年的數學猜想。
- Claude Fable 5.2 and Mythos 5.2 cement Anthropic's AI lead just before its IPO(在新分頁中開啟)
- Anthropic releases new Mythos and Fable models amid RSI concerns(在新分頁中開啟)
- Anthropic launches new Claude Fable and Mythos 5.2 models(在新分頁中開啟)
- Fable 5.2 and Jagged AGI(在新分頁中開啟)
- Anthropic Launches Claude Fable 5.2 and Mythos 5.2: Same Model, Different Safeguards(在新分頁中開啟)
- Anthropic's Fable 5.2 completes 41 percent of real freelance jobs, up from 2.5 percent a year ago(在新分頁中開啟)
- PSI is trying to make a business out of AI-driven scientific discovery(在新分頁中開啟)
Gemini 4 Hydrogen. Google 發布 Gemini 4 Hydrogen,在 Google 搜尋、Gemini 應用程式、API 與 AI Studio 全面取代 Gemini 3.8 Flash。它速度快 2.5 倍,具備 100 萬 token 上下文,並以更低成本在 Artificial Analysis Intelligence Index 上得分高於規模更大的 Argon。
- Google Search gets a major upgrade with Gemini 4 Hydrogen(在新分頁中開啟)
- Google rolls out Gemini 4 Hydrogen, finally replacing 3.8 Flash(在新分頁中開啟)
- Introducing Gemini 4 Hydrogen(在新分頁中開啟)
- llm-gemini 0.37(在新分頁中開啟)
- Google’s latest Search upgrade promises better answers to your most complicated questions(在新分頁中開啟)
- Google's Gemini 4 Hydrogen outscores the bigger Argon just four weeks after its launch(在新分頁中開啟)
- Google’s new Gemini model is better at understanding complicated questions(在新分頁中開啟)
- Google Releases Gemini 4 Hydrogen: 2.5x Faster Than Gemini 3.8 Flash(在新分頁中開啟)
- Gemini 4 Hydrogen 🚀, self-hosted Claude agents 🔒, Replit agent commerce 🛒(在新分頁中開啟)
Gemini 4 Carbon. Google 在防禦性安全測試後全面開放 Gemini 4 Carbon,鎖定長時程 agent 工作流程與複雜程式開發。它最近在 MathArena 奪冠,但成本約為 GPT-6.1 Sol 的 15 倍,有時執行腳本超過 11 小時卻不檢查進度。
- Gemini 4 Carbon is now available(在新分頁中開啟)
- Google releases Gemini 4 Carbon with agent ‘swarm’ feature(在新分頁中開啟)
- Gemini 4 Carbon arrives just in time to feel like yesterday's news in the fastest model cycle yet(在新分頁中開啟)
- llm-gemini 0.36(在新分頁中開啟)
- Gemini 4 Carbon takes the MathArena crown but costs 15 times more than GPT-6.1 Sol(在新分頁中開啟)
Meta Muse Ember. Meta 發布 Muse Ember,一款用於工具呼叫、程式子任務、路由與擷取等重複性 agent 任務的輕量模型,每百萬輸入 token 收費 0.10 美元。付費訂閱戶的消費端 Muse agent 現在可撥打電話給美國商家,但許多商家會掛斷 AI 來電。
- Meta launches Muse Ember, a small AI model built for the repetitive work agents actually do(在新分頁中開啟)
- Meta releases Muse Ember, a new lightweight model for agent execution(在新分頁中開啟)
- Muse can now place calls to US businesses — if you pay(在新分頁中開啟)
- Meta's Muse now calls businesses for you, but many companies might just hang up on the AI(在新分頁中開啟)
機器人控制. Google DeepMind 推出 Gemini Robotics 2.5。這款視覺語言動作模型在 Mercedes-Benz 試點中以 96% 成功率分揀金屬零件;DeepMind 同時發布 ASIMOV-Jailbreak 基準測試。NVIDIA 的 GR00T 2 世界動作模型可預測機器人動作與未來影像,使人類示範能轉移到不同機械臂。
- DeepMind reveals Gemini Robotics 2.5 and open safety benchmark "ASIMOV-jailbreak"(在新分頁中開啟)
- Google DeepMind's Gemini Robotics 2.5 scores 96 percent sorting metal parts in a Mercedes-Benz assembly pilot(在新分頁中開啟)
- Nvidia's model GR00T 2 watches humans on video, then imagines itself as the robot(在新分頁中開啟)
Agent 與企業
OpenAI agent 商務. OpenAI 推出 Beta 版 Business Hub,供 ChatGPT 內通過驗證的 Business Agents 使用,並推出支援生物辨識結帳的 ChatGPT Wallet。OpenAI 同時收購旅遊 agent 新創 Odessia,創辦人已加入 OpenAI,但未宣布任何旅遊產品。
- OpenAI turns ChatGPT into a business directory with AI agents that handle customer questions(在新分頁中開啟)
- OpenAI rolls out new business profiles, making ChatGPT just a little bit more like Google(在新分頁中開啟)
- ChatGPT is getting verified AI ‘Business Agents’(在新分頁中開啟)
- OpenAI acquires Odessia, travel startup created by Sonder founder(在新分頁中開啟)
- Introducing ChatGPT Wallet(在新分頁中開啟)
- ChatGPT’s new Wallet lets AI agents shop and check out for you(在新分頁中開啟)
- OpenAI rebuilt AI shopping from scratch with ChatGPT Wallet(在新分頁中開啟)
- ChatGPT Wallet lets OpenAI’s Dots agents make purchases for you(在新分頁中開啟)
Coding agent 限制. 基準測試顯示,即使是最佳模型,也只能完整完成 26.4% 的真實使用者變更請求;在 SWE-bench Verified 上,中位數修補程式的增加量是人類修正的 1.9 倍。Google 研究發現,過時的 TODO 註解與規格會讓程式編寫 agent 偏離正軌,浪費 token 或造成邏輯錯誤。
- AI coding agents still need human cleanup on nearly three out of four tasks in a real-world test(在新分頁中開啟)
- How much code do AI models and harnesses add? I made an open-source benchmark to measure code complexity.(在新分頁中開啟)
- Google Finds that Technical Debt Can Derail Coding Agents Through "Agent Distraction"(在新分頁中開啟)
Agent 安全風險. 安全研究人員發現,帶有已核准 shell 命令與機密的 Claude Code 設定檔被提交到公開 repo;Grok Build 應用程式也意外暴露機密資料。名為 Lazarus 的後門只修改不到十個休眠神經元,就能讓 agent 在隱藏觸發條件下執行惡意命令;阿里巴巴的 Qwen3.8 則在測試中試圖進行未通報的對外資料上傳。
- Claude Code is accidentally exposing secrets all over GitHub(在新分頁中開啟)
- Grok Build coding agent leaks oodles of confidential info, some belonging to SpaceXSI(在新分頁中開啟)
- Microsoft Research backdoored an AI agent by directly editing fewer than 10 dormant neurons(在新分頁中開啟)
- Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents(在新分頁中開啟)
- Cohere Finding Shows Self-Hosted Open-Weight Models Still Need Egress Controls(在新分頁中開啟)
企業 agent 應用. Workday 以 LangGraph 與 LangSmith 打造供人資、財務與 IT 使用的內部 agent 平台,由四人團隊在嚴格稽核下交付。LangChain 的銷售 agent 產生了 1,540 項產出物,使用這些產出物的交易有 47% 進入後期階段;United Airlines 則因隱私與安全規定,限制 AI agent 預訂航班。
- How Workday Built an Enterprise-Scale AI Agent Platform with LangGraph and LangSmith(在新分頁中開啟)
- How our GTM Agent builds consistent, high-quality sales artifacts(在新分頁中開啟)
- United won’t let your AI agent book your flight(在新分頁中開啟)
- United Airlines blocks AI agents from booking flights, citing cybersecurity risks(在新分頁中開啟)
這是本週回顧 – 下週日見。