Xi’s White House State Dinner Quickly Becoming an AI Meet Up
· The Aigentic · Morning Brief
Bloomberg reports Satya Nadella plans to attend this week’s White House state dinner for Xi Jinping, joining Sam Altman, Jensen Huang, Tim Cook, and Qualcomm’s Cristiano Amon — with officials also naming Bezos, Musk, and Pichai among expected guests — turning the summit pageantry into an AI-industry stage.
The Guardian reports British Columbia sued OpenAI and Sam Altman in San Francisco federal court, saying the February Tumbler Ridge school massacre might have been stopped if the company had warned local police after ChatGPT chats about gun violence were flagged — safety staff recommended calling law enforcement, leadership overruled them, the account was deactivated, and the shooter opened a new one and kept planning; the province wants damages plus a court order on how OpenAI handles violence-related chats.
Semafor reports OpenAI on Monday urged the U.S. to lead an international coalition on frontier AI standards as UN General Assembly week opens, framing voluntary technical benchmarks through the network of AI safety institutes while Washington pursues tentative China outreach and a White House that has so far resisted domestic safety mandates.
Reuters reports Anthropic and OpenAI urged Australia to rethink its ban on training AI on local creative content, proposing a narrow, conditional path after Canberra ruled out a broad copyright exemption, and tying the ask to data-center offtakes already underway with NextDC in Sydney and a Queensland partnership for Anthropic.
Reuters reports Alibaba, at Apsara in Hangzhou, said its Qwen line is aiming at models with 5–10 trillion parameters — versus 2.4 trillion for Qwen 3.8 Max — and unveiled the Zhenwu V900 AI chip from T-Head, pitched as China’s most powerful and due for mass production in Q1 2027, while CEO Eddie Wu set a 20 GW global cloud capacity target by 2032; shares jumped about 5%.
TechCrunch reports Google opened $899 Googlebook preorders — Android laptops with Magic Cursor, Rambler dictation cleanup, and Gemini baked in — shipping early October as a Chromebook-successor bet for schools and consumers, with flagship machines from Acer, ASUS, Dell, HP, and Lenovo.
Xi’s White House State Dinner Includes an AI Summit’s Guest List
Bloomberg reports Microsoft CEO Satya Nadella plans to attend this week’s White House state dinner for Chinese President Xi Jinping, joining OpenAI’s Sam Altman, Nvidia’s Jensen Huang, Apple’s Tim Cook, and Qualcomm’s Cristiano Amon. Officials have also named Jeff Bezos, Elon Musk, and Google’s Sundar Pichai among expected guests. The dinner turns summit pageantry into an AI-industry stage as U.S.–China talks already float a formal AI dialogue and incident-notification mechanism.
That matters because the people who build and buy frontier compute will be in the room when Trump and Xi set the tone for AI diplomacy — optics that can move chip policy and export narratives overnight.
The one move is to watch which talking points survive the dinner briefings: cooperation language, chip sales, or nothing at all. Read Bloomberg
The safety team said call the police
The Guardian reports British Columbia sued OpenAI and Sam Altman in San Francisco federal court over the February Tumbler Ridge school shooting that killed nine people, primarily children. The province alleges the massacre could have been prevented if OpenAI had warned local police after the shooter’s ChatGPT conversations about gun violence were flagged. Safety team members recommended contacting law enforcement; Altman and other leadership overruled them, the suit says, citing whistleblowers who spoke to the Wall Street Journal. The account was deactivated, but the shooter opened a new one and kept planning. The province seeks damages for recovery costs — including a replacement school and mental health care — plus a court order forcing changes to how OpenAI handles violence-related chats. Altman had apologized and promised reforms that the suit says never arrived.
That matters because a Canadian province is asking a U.S. federal court to treat an internal safety override — and the easy path to a new account — as actionable negligence after a school massacre, not just a product-policy debate.
The one move is to watch whether the court lets the case reach discovery on who overruled the flag and what the “imminent and credible risk” bar actually meant in practice. Read The Guardian
America should run the standards table
Semafor reports OpenAI on Monday called for the United States to lead an international coalition coordinating global AI standards as UN General Assembly week opens. The pitch lands beside tentative U.S.–China outreach on AI — including talk of a deconfliction-style hotline — and a White House that has so far been reluctant to shape domestic AI safety mandates, aligning publicly with industry bulls.
That matters because the same company facing a province lawsuit over a missed police call is asking Washington to own the global rulebook, while China talks stay limited and domestic mandates stay soft.
The one move is to separate coalition branding from enforceable reporting and evaluation requirements before treating the pitch as a safety win. Read Semafor
Training wants a narrow door Down Under
Reuters reports Anthropic and OpenAI urged Australia to rethink a ban on training AI models on the country’s creative content. In submissions to a parliamentary inquiry, both acknowledged Canberra has ruled out a broad copyright exemption and floated a narrow, conditional approval path — including possible investment conditions to support Australian creators. OpenAI has an offtake deal with NextDC for a Sydney data center; Anthropic was named last week as a partner for a Queensland facility. The Joint Select Committee is due to report in November.
That matters because the labs are tying copyright relief to capital already committed on Australian soil — making training rights a condition of the infra story, not a separate culture war.
The one move is to track whether the November report opens any conditional training path or doubles down on the ban. Read Reuters
Five to ten trillion, plus a domestic chip
Reuters reports Alibaba said at its Apsara conference in Hangzhou that the Qwen team plans models with 5 trillion to 10 trillion parameters — versus 2.4 trillion for flagship Qwen 3.8 Max — and unveiled the Zhenwu V900 AI chip from T-Head, pitched as China’s most powerful, with clusters of up to 500,000 chips and mass production targeted for Q1 2027. CEO Eddie Wu also set a goal for Alibaba Cloud’s global data-center capacity to surpass 20 gigawatts by 2032. Hong Kong-listed shares jumped about 5%.
That matters because a Chinese hyperscaler is publishing a concrete scale-up roadmap for both models and silicon while U.S. export curbs keep Nvidia alternatives scarce.
The one move is to treat the V900 ship date and the 20 GW target as the real meters — parameter counts alone are marketing until capacity and chips land. Read Reuters
They want you to buy a new laptop for Gemini
TechCrunch reports Google opened preorders for Googlebook at $899 — Android laptops with Magic Cursor, Rambler dictation cleanup, vibe-coded widgets, and Gemini baked in — shipping October 4 in the U.S. and October 5 in several other markets. Flagship machines come from Acer, ASUS, Dell, HP, and Lenovo, with up to 10 years of updates and 12 months of Google AI Pro included. The bet is that Gemini on device is enough to pull Chromebook users — including a large K-12 installed base — onto a new hardware line.
That matters because Google is trying to make the laptop itself the Gemini distribution channel, not just another place to open a chat tab.
The one move is to ask whether Magic Cursor and Rambler justify a hardware refresh — or whether agents that already drive the browser make the new machine optional. Read TechCrunch
Grok 4.7 joins The Aigentic models report
xAI’s Grok 4.7 is now on The Aigentic models board — same $2 / $6 list as Grok 4.6 on the standard band, 500k context, AA Index 46 on v4.3.2 at xhigh. See every tracked model, price, and score in one place. View all models
Deals Desk
New compute and power tape from the tracker, newest first.
2026-09-21 — New Era Energy & Digital × Vistra/Luminant — 20-year PPA, minimum 200 MW up to 207 MW for Phase 1 of the Texas Critical Data Center in Odessa; power from the adjacent Vistra Odessa gas plant, expected Q3 2027. Source: GlobeNewswire
2026-09-21 — Sharon AI × GreenSquare — Long-term SYD1 Stage 1 capacity in Western Sydney for about 8,200 NVIDIA Blackwell Ultra GPUs, operations targeted Q4 2026. Source: TechPartner
2026-09-21 — Unnamed West Texas DC operator × Kodiak Gas Services — Six-year 76 MW behind-the-meter baseload power agreement; deploy Q4 2026–Q1 2027. Source: Kodiak IR
Full tracker → https://www.theaigentic.com/deals
Aigentic Exclusive
Hands On: Jev Wants to Make AI Decisions Almost Free. Here’s Where We’d Put It to Work.
TypeSafe’s new model brings an intriguing mix of speed, low prices, and tightly defined answers to business automation. The opportunity is real. So is the work required to make those answers trustworthy.
Watch
This morning’s Watch opens on the state of the AI policy debate, then stress-tests Grok 4.7 twice — release scorecard and Grok Bot hands-on — before Claude Projects, a five-minute Jev primer, and a builder how-to for decision-only models.
The AI Daily Brief — State of the AI Debate
NLW packs overnight policy tape into one pass — Trump’s AI Force push, state kill-switch crackdowns, and China talks — the cleanest news-analysis open for a Tuesday Brief.
The one move is naming which of those three threads already changed a decision on your roadmap this week.
Matthew Berman — Did Elon catch up? Grok 4.7 is here
Matthew Berman covers the Grok 4.7 release with testing — where the model lands on the scorecard versus prior Grok versions, and what still separates it from the frontier pack.
The one move is checking the same tasks you already run on Claude or GPT before swapping a default.
Alex Finn — Grok 4.7 inside Grok Bot
Alex Finn puts Grok 4.7 to work inside Grok Bot — a hands-on workflow angle that complements Berman’s release read and shows what the model does when it has tools and memory in the loop.
The one move is picking one agent workflow to try in Grok Bot before you rewrite your routing rules.
Riley Brown — NEW Claude Projects changes everything
Riley Brown walks through Claude Projects with shared memory and artifacts across agent threads — including three real business projects spun up in 48 hours — and why persistent project context changes how you staff Claude Code.
The one move is converting one recurring client or product thread into a Project before starting the next blank chat.
Fireship — An ex-OpenAI researcher just deleted language from the LLM
Fireship packs TypeSafe’s Jev — a System 1 judgment model that skips language generation — into a dense five minutes, framing why an ex-OpenAI researcher’s “deleted language” bet matters for the decision-only wave.
The one move is separating judgment classifiers from chat models before you price the next automation seat.
Cole Medin — How to actually use Jev
Cole Medin shows where decision-only classifiers like Jev actually belong in apps — a builder how-to after the hype cycle, not another SOTA reaction — and what you still need a language model to do.
The one move is mapping one high-volume yes/no gate in your product to a judgment model before wrapping it in a chat UI.