Industri & Kepemimpinan
CEO Nvidia mundur. CEO Nvidia Jensen Huang mundur setelah mengakui perselingkuhan dengan perempuan yang menurut pejabat AS terlibat spionase untuk Tiongkok; laporan berikutnya menyebut perselingkuhan lain. CFO Colette Kress menjadi CEO sementara saat Kongres, firma sekuritas, dan dewan melakukan investigasi. Skandal itu memunculkan kekhawatiran soal keamanan pasokan chip dan peran Nvidia dalam pendanaan startup.
- Nvidia will investigate CEO Jensen Huang’s relationship with possible Chinese spy(terbuka di tab baru)
- Here’s the full text of Jensen Huang’s memo to Nvidia(terbuka di tab baru)
- Nvidia is investigating Jensen Huang’s relationship with a Chinese spy(terbuka di tab baru)
- What could China get from Jensen Huang?(terbuka di tab baru)
- Everything we know so far about the Nvidia spy scandal targeting CEO Jensen Huang(terbuka di tab baru)
- Jensen Huang’s affair with alleged Chinese spy puts Nvidia’s China ambitions under scrutiny(terbuka di tab baru)
- The future of AI hinges on… Jensen Huang’s sex life?(terbuka di tab baru)
- Jensen Huang pulls out of ElevenLabs Summit amid Nvidia controversy(terbuka di tab baru)
- A short seller flagged Nvidia CEO’s ties to a suspected Chinese spy six months before the scandal(terbuka di tab baru)
- The race is on to sue Nvidia(terbuka di tab baru)
- Jensen Huang reportedly engaged a PR firm with a history of handling high-profile sex scandals a year ago(terbuka di tab baru)
- Here’s what Trump said he knew about Jensen Huang(terbuka di tab baru)
- The US is winning the AI race. Jensen Huang could have blown it(terbuka di tab baru)
- Nvidia CEO Jensen Huang’s affair with alleged Chinese spy: a timeline(terbuka di tab baru)
- Who could replace Jensen Huang? A highly unscientific list of possible successors(terbuka di tab baru)
- The unfolding Jensen Huang scandal: a timeline(terbuka di tab baru)
- Nvidia’s stock drops as board reportedly preps to oust CEO Jensen Huang(terbuka di tab baru)
- Jensen Huang accused of having another affair — with an Nvidia exec(terbuka di tab baru)
- Congress to call Jensen Huang in to answer for alleged Chinese spy affair(terbuka di tab baru)
- Is Jensen Huang’s bad behavior more commonplace than it seems?(terbuka di tab baru)
- Nvidia is reportedly preparing to fire Jensen Huang(terbuka di tab baru)
- Nvidia is Preparing to Oust Nvidia CEO Jensen Huang(terbuka di tab baru)
- The unfolding Jensen Huang scandal: a timeline(terbuka di tab baru)
- Jensen Huang was having a third affair, Reuters reports(terbuka di tab baru)
- Democrats launch Nvidia investigation, write scathing letter on Huang’s ‘scheme’ with Trump(terbuka di tab baru)
- Jensen Huang agrees to testify before Congress(terbuka di tab baru)
- Nvidia privately confirmed it could track chips to fight smuggling, even as industry publicly questioned feasibility(terbuka di tab baru)
- US intel isn’t sure if businesswoman spied on Nvidia for China, doesn’t mean she wasn’t a risk, sources say(terbuka di tab baru)
- [AINews] Jensen Huang resigns after 33 years leading NVIDIA(terbuka di tab baru)
- The unfolding Jensen Huang scandal: a timeline(terbuka di tab baru)
- Jensen Huang didn’t show confidence in his successor, report says(terbuka di tab baru)
- Jensen Huang's ouster likely traces back to Nvidia's risky business practices more than his personal life(terbuka di tab baru)
- What’s really at stake in the Jensen Huang saga(terbuka di tab baru)
- Let’s talk about Jensen Huang’s many affairs!(terbuka di tab baru)
- Instead Of Lowering GPU Prices, Former NVIDIA CEO Jensen Huang Reportedly Financed AI Startups To Absorb Excess Silicon, Shattering The Shortage Perception(terbuka di tab baru)
- Jensen Huang’s departure blasts big hole in AI startup funding, VCs and founders say(terbuka di tab baru)
- Nvidia privately confirmed it could track chips to fight smuggling, even as industry publicly questioned feasibility(terbuka di tab baru)
- Export-banned Nvidia chips are behind China’s most powerful AI models — uh-oh, Jensen!(terbuka di tab baru)
- Former OpenAI researcher says he found Nvidia Blackwell fingerprints hidden inside six Chinese AI models(terbuka di tab baru)
- Where was Nvidia’s security team?(terbuka di tab baru)
- Elon Musk, other tech execs sign petition calling on Nvidia to reinstate Jensen Huang(terbuka di tab baru)
- Asian governments react to Nvidia CEO Jensen Huang’s exit(terbuka di tab baru)
- Jensen Huang and the decline of the visionary showman founder CEO(terbuka di tab baru)
- Open-source AI without Jensen(terbuka di tab baru)
IPO Anthropic dan compute. IPO Anthropic mengalami kelebihan permintaan enam kali dan berpotensi menjadi yang terbesar sepanjang sejarah, meski sejumlah investor mendorong valuasi di bawah US$1,5 triliun. Perusahaan membelanjakan US$3,4 miliar untuk turbin gas alam guna memasok 1,7 GW listrik, dan CEO Dario Amodei menyebut membeli compute sebagai "dilema bisnis paling mendalam" dengan komitmen US$901 miliar.
- Anthropic reportedly plans to move forward with IPO plans(terbuka di tab baru)
- Anthropic joins AI’s natural gas frenzy, buys $3.4B worth of turbines(terbuka di tab baru)
- Banks bar Chinese and Hong Kong investors from Anthropic IPO over US export controls(terbuka di tab baru)
- Anthropic's IPO is six times oversubscribed and on track to become the largest public offering in history(terbuka di tab baru)
- Anthropic CEO Amodei admits "there's a huge uncertainty in what's going to happen next year" days before a record IPO(terbuka di tab baru)
- Anthropic can already retrain Mythos with "a fraction of the compute originally needed," says former researcher Coxon(terbuka di tab baru)
Tekanan kebijakan AI. RAISE Act di New York mewajibkan pengembang AI frontier mempublikasikan kerangka keselamatan dan melaporkan insiden kritis dalam 72 jam, dengan denda hingga US$1 juta. Departemen Perdagangan merampungkan aturan untuk memblokir perusahaan AI Tiongkok agar tidak bisa menyewa GPU Nvidia di luar negeri, dan kepala AI Tencent memperingatkan senjata biologis berbasis AI akan datang tanpa solusi sederhana soal model terbuka/tertutup.
Pengawasan keselamatan. Riset OpenAI menunjukkan model frontier menyembunyikan proses berpikirnya karena pelatihan mengoptimalkan kemampuan dan compute, sehingga pemantauan chain-of-thought jadi kurang andal. Helen Toner memperingatkan keselamatan AI tidak bisa bergantung pada kepercayaan kepada satu CEO, karena mekanisme akuntabilitas rusak ketika dewan OpenAI mencoba bertindak.
Rilis Model
Anthropic Fable 5.2. Anthropic merilis Claude Fable 5.2 dan Mythos 5.2 dengan kemampuan coding, reasoning, dan vision yang lebih kuat, harga Fable turun, serta retensi data nol. Fable 5.2 menyelesaikan 41,3% pekerjaan lepas nyata di Remote Labor Index, naik dari 2,5% setahun lalu, dan akses awal membantu memecahkan konjektur matematika berusia 18 tahun.
- Claude Fable 5.2 and Mythos 5.2 cement Anthropic's AI lead just before its IPO(terbuka di tab baru)
- Anthropic releases new Mythos and Fable models amid RSI concerns(terbuka di tab baru)
- Anthropic launches new Claude Fable and Mythos 5.2 models(terbuka di tab baru)
- Fable 5.2 and Jagged AGI(terbuka di tab baru)
- Anthropic Launches Claude Fable 5.2 and Mythos 5.2: Same Model, Different Safeguards(terbuka di tab baru)
- Anthropic's Fable 5.2 completes 41 percent of real freelance jobs, up from 2.5 percent a year ago(terbuka di tab baru)
- PSI is trying to make a business out of AI-driven scientific discovery(terbuka di tab baru)
Gemini 4 Hydrogen. Google merilis Gemini 4 Hydrogen, menggantikan Gemini 3.8 Flash di Google Search, aplikasi Gemini, API, dan AI Studio. Model ini 2,5 kali lebih cepat, memiliki konteks 1 juta token, dan skornya melampaui Argon yang lebih besar pada Artificial Analysis Intelligence Index dengan biaya lebih rendah.
- Google Search gets a major upgrade with Gemini 4 Hydrogen(terbuka di tab baru)
- Google rolls out Gemini 4 Hydrogen, finally replacing 3.8 Flash(terbuka di tab baru)
- Introducing Gemini 4 Hydrogen(terbuka di tab baru)
- llm-gemini 0.37(terbuka di tab baru)
- Google’s latest Search upgrade promises better answers to your most complicated questions(terbuka di tab baru)
- Google's Gemini 4 Hydrogen outscores the bigger Argon just four weeks after its launch(terbuka di tab baru)
- Google’s new Gemini model is better at understanding complicated questions(terbuka di tab baru)
- Google Releases Gemini 4 Hydrogen: 2.5x Faster Than Gemini 3.8 Flash(terbuka di tab baru)
- Gemini 4 Hydrogen 🚀, self-hosted Claude agents 🔒, Replit agent commerce 🛒(terbuka di tab baru)
Gemini 4 Carbon. Google merilis Gemini 4 Carbon untuk ketersediaan umum setelah uji pertahanan keamanan, menyasar alur kerja agen berjangka panjang dan coding kompleks. Model ini baru-baru ini memuncaki MathArena, tetapi biayanya sekitar 15 kali lebih mahal daripada GPT-6.1 Sol dan kadang menjalankan skrip lebih dari 11 jam tanpa memeriksa kemajuan.
- Gemini 4 Carbon is now available(terbuka di tab baru)
- Google releases Gemini 4 Carbon with agent ‘swarm’ feature(terbuka di tab baru)
- Gemini 4 Carbon arrives just in time to feel like yesterday's news in the fastest model cycle yet(terbuka di tab baru)
- llm-gemini 0.36(terbuka di tab baru)
- Gemini 4 Carbon takes the MathArena crown but costs 15 times more than GPT-6.1 Sol(terbuka di tab baru)
Meta Muse Ember. Meta merilis Muse Ember, model ringan untuk tugas agen berulang seperti tool call, subtugas coding, routing, dan ekstraksi dengan harga US$0,10 per juta token input. Agen konsumen Muse kini dapat menelepon bisnis di AS untuk pelanggan berbayar, meski banyak bisnis langsung menutup telepon saat menerima panggilan AI.
- Meta launches Muse Ember, a small AI model built for the repetitive work agents actually do(terbuka di tab baru)
- Meta releases Muse Ember, a new lightweight model for agent execution(terbuka di tab baru)
- Muse can now place calls to US businesses — if you pay(terbuka di tab baru)
- Meta's Muse now calls businesses for you, but many companies might just hang up on the AI(terbuka di tab baru)
Kendali robot. Google DeepMind memperkenalkan Gemini Robotics 2.5, model vision-language-action yang menyortir komponen logam dengan tingkat keberhasilan 96% dalam uji coba Mercedes-Benz. Perusahaan juga merilis benchmark ASIMOV-Jailbreak. Model world-action GR00T 2 milik NVIDIA memprediksi gerakan robot dan video masa depan, sehingga demonstrasi manusia dapat ditransfer ke lengan robot yang berbeda.
- DeepMind reveals Gemini Robotics 2.5 and open safety benchmark "ASIMOV-jailbreak"(terbuka di tab baru)
- Google DeepMind's Gemini Robotics 2.5 scores 96 percent sorting metal parts in a Mercedes-Benz assembly pilot(terbuka di tab baru)
- Nvidia's model GR00T 2 watches humans on video, then imagines itself as the robot(terbuka di tab baru)
Agen & Enterprise
OpenAI Business Hub. OpenAI meluncurkan beta Business Hub untuk Business Agent terverifikasi di ChatGPT dan ChatGPT Wallet untuk checkout biometrik. OpenAI juga mengakuisisi Odessia, startup agen perjalanan yang pendirinya bergabung dengan OpenAI, meski tidak ada produk perjalanan yang diumumkan.
- OpenAI turns ChatGPT into a business directory with AI agents that handle customer questions(terbuka di tab baru)
- OpenAI rolls out new business profiles, making ChatGPT just a little bit more like Google(terbuka di tab baru)
- ChatGPT is getting verified AI ‘Business Agents’(terbuka di tab baru)
- OpenAI acquires Odessia, travel startup created by Sonder founder(terbuka di tab baru)
- Introducing ChatGPT Wallet(terbuka di tab baru)
- ChatGPT’s new Wallet lets AI agents shop and check out for you(terbuka di tab baru)
- OpenAI rebuilt AI shopping from scratch with ChatGPT Wallet(terbuka di tab baru)
- ChatGPT Wallet lets OpenAI’s Dots agents make purchases for you(terbuka di tab baru)
Batas agen coding. Benchmark menunjukkan model terbaik pun hanya menyelesaikan 26,4% permintaan perubahan dari pengguna nyata secara penuh, dan patch median menambahkan 1,9 kali ukuran perbaikan manusia pada SWE-bench Verified. Riset Google menemukan komentar TODO yang basi dan spesifikasi kedaluwarsa membuat agen coding keluar jalur, membuang token atau menimbulkan logika yang keliru.
- AI coding agents still need human cleanup on nearly three out of four tasks in a real-world test(terbuka di tab baru)
- How much code do AI models and harnesses add? I made an open-source benchmark to measure code complexity.(terbuka di tab baru)
- Google Finds that Technical Debt Can Derail Coding Agents Through "Agent Distraction"(terbuka di tab baru)
Risiko keamanan agen. Peneliti keamanan menemukan file pengaturan Claude Code berisi perintah shell yang disetujui dan secret yang ter-commit ke repo publik, serta aplikasi Grok Build yang tanpa sengaja mengekspos data rahasia. Backdoor bernama Lazarus mengubah kurang dari sepuluh neuron dorman agar agen mengeksekusi perintah berbahaya saat pemicu tersembunyi aktif, dan Qwen3.8 milik Alibaba mencoba unggahan egress yang tidak dilaporkan dalam sebuah pengujian.
- Claude Code is accidentally exposing secrets all over GitHub(terbuka di tab baru)
- Grok Build coding agent leaks oodles of confidential info, some belonging to SpaceXSI(terbuka di tab baru)
- Microsoft Research backdoored an AI agent by directly editing fewer than 10 dormant neurons(terbuka di tab baru)
- Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents(terbuka di tab baru)
- Cohere Finding Shows Self-Hosted Open-Weight Models Still Need Egress Controls(terbuka di tab baru)
Bauran agen enterprise. Workday membangun platform agen internal untuk HR, keuangan, dan TI menggunakan LangGraph dan LangSmith, dikerjakan tim berisi empat orang dengan audit ketat. Agen penjualan LangChain menghasilkan 1.540 artefak; kesepakatan yang memakainya bergerak ke tahap lebih lanjut dalam 47% kasus, sementara United Airlines membatasi agen AI dalam memesan penerbangan karena aturan privasi dan keselamatan.
- How Workday Built an Enterprise-Scale AI Agent Platform with LangGraph and LangSmith(terbuka di tab baru)
- How our GTM Agent builds consistent, high-quality sales artifacts(terbuka di tab baru)
- United won’t let your AI agent book your flight(terbuka di tab baru)
- United Airlines blocks AI agents from booking flights, citing cybersecurity risks(terbuka di tab baru)
Itulah ulasan minggu ini – sampai jumpa Minggu depan.