Friday, 18 September 2026 · Daily Briefing

Pulse AI Briefing

King Charles convened OpenAI, Anthropic, DeepMind and Nvidia at a private summit in Scotland on the same week two of those labs published their own failure logs — while Sam Altman and Jensen Huang took seats at the White House table for Xi Jinping. The safety mood and the power dinner are now running in parallel, and nobody has reconciled them.

The signal

The labs' safety warnings become the industry's political weather

On 17 September King Charles hosted leaders from OpenAI, Anthropic, Google DeepMind and Nvidia at Dumfries House in Ayrshire, the headquarters of the King's Foundation, with the UK's AI minister and roughly thirty invited guests. The palace framed it as a conversation about developing AI in ways that benefit society; the King, who opened the world's first AI safety summit in 2023, urged executives to keep the technology in the service of humanity. The same week, OpenAI published a framework disclosing six incidents of model misalignment, Anthropic detailed disruption of Claude misuse across seven harm areas, and both Brussels and the UN secretary-general publicly pressed the case for pacing the frontier. Yet on 24 September Sam Altman and Jensen Huang will sit at a White House state dinner for Xi Jinping, where AI is the principal agenda item and the guest list is the tell. Two tracks are now running at once: the voluntary-disclosure track, where labs publish their own failures, and the deal track, where compute diplomacy decides what gets sold. Watch which one sets the rules.

Seven stories

Today's Briefing

Policy · 17 September

The King convenes the labs that say they should slow down

King Charles brought together representatives of OpenAI, Anthropic, Google DeepMind and Nvidia at Dumfries House in Ayrshire, along with the UK's AI minister and about thirty invited guests, to discuss how AI can be developed and deployed in ways that benefit society. The King, who addressed the world's first AI safety summit in 2023, asked executives to align the technology with his "philosophy of Harmony" and to keep it under human control. OpenAI sent chief financial officer Sarah Friar. The gathering came as the leaders of the frontier labs themselves have publicly warned that capability improvement should be paced — a reversal that hands the safety argument to institutions rather than leaving it inside the companies.

Source: NPRPrimary: Associated Press

Politics · 17 September

Altman and Huang take seats at the Xi state dinner

OpenAI chief executive Sam Altman and Nvidia chief executive Jensen Huang are planning to attend the White House state dinner for Chinese president Xi Jinping on 24 September, with Apple executive chairman Tim Cook also confirmed by ABC News. AI is expected to dominate the meeting between the two presidents. Huang's attendance underscores Nvidia's position in the US-China technology dialogue: he travelled to China with Trump earlier this year and has argued publicly that AI doomsday scenarios are not grounded in science. Anthropic had not said whether Dario Amodei would attend. The dinner closes a week in which the administration dismissed domestic AI warnings while the labs urged caution.

Source: ABC NewsPrimary: Reuters

Hardware · 17 September

Huawei pulls the Ascend 960DT forward by three quarters

Huawei will launch the Ascend 960DT in the first quarter of 2027 rather than the fourth, and the 960PR in Q3 2027, rotating and acting chairman David Wang told the Huawei Connect conference in Shanghai — three quarters ahead of the 2025 roadmap. Each 960DT delivers 2 PFLOPS at FP8 and 4 PFLOPS at FP4 with up to 288GB of HBM at 9.6 TB/s. Huawei also unveiled a 4,096-NPU super node rated at 8 EFLOPS FP8, using NPO packaging that replaces roughly 48,000 800G optical modules with 5,500 Hi-ONE units and cuts more than 550 kW. The cadence continues to Ascend 970 in 2028 and 980 in 2029. Wang said Chinese demand exceeds Huawei's capacity, limiting overseas sales — and analyst Rui Ma noted the announced super node is far smaller than the 15,488-chip Atlas 960 SuperPoD Huawei previously described.

Source: ReutersPrimary: TechCrunch

Security · 17 September

Anthropic says two Chinese labs sold Claude as their own

Anthropic's September threat-intelligence report, covered widely on 17 September, alleges that Moonshot AI — maker of the Kimi family — silently forwarded customer requests to Claude and displayed Claude's answers to users who believed they were talking to a Moonshot model, and that DeepSeek did the same. Anthropic says Moonshot relayed almost 300,000 customer requests in a single ten-day stretch through a network of 5,380 accounts it describes as fraudulent, mostly traced to Singapore and Japan, and attributes more than 23 million exchanges to the family of activity between May and July 2026. DeepSeek is said to have tagged traffic arriving through Claude Code, the Claude Agent SDK and OpenCode, then rerouted selected sessions to Claude Opus. Anthropic frames it as illicit distillation with a live routing twist; neither company has responded publicly. The 154-page report covers seven harm areas, from cyber operations to biological misuse.

Source: South China Morning PostPrimary: Anthropic

Product · 16 September

Snap prices the AR bet at $2,195 and ships an anticipatory assistant

Snap launched its first consumer augmented-reality glasses at $2,195 and, with them, Specs Intelligence — an assistant the company calls anticipatory, built to learn goals, priorities, relationships and routines from apps users connect, including Messages, Gmail and Slack. It runs on the glasses but also as standalone Mac and iPhone apps, routing tasks such as checking a changed gate or buying cinema tickets with a confirmation step before purchase. Chief executive Evan Spiegel positioned the launch against AI pessimism, arguing computers "should make us more important, not less". A limited US preview is live through the Specs app on iOS; the full early-access version is invite-only and initially tied to the Mac app. No ship date was given, with a Los Angeles pop-up planned for October.

Source: TechCrunchPrimary: Engadget

Platform · 16 September

Google hands the smart home to third-party agents

Google opened early access to a Home MCP server, letting any agent that can call Model Context Protocol tools — Claude, ChatGPT, Google Antigravity, Hermes and OpenClaw among those named — discover devices, read state and event history, and execute authorised actions across the Google Home ecosystem. Access is rolling out in English to Google Home Premium Advanced subscribers in the United States at $20 a month, and setup requires a Google Cloud project, an OAuth consent screen and a compatible MCP client. Google says sensitive actions are banned outright, call frequency is throttled and extra consent is required for sensitive data. The residual risk it acknowledges is the obvious one: an agent that misreads context while operating entirely inside its granted permissions.

Source: TechCrunch

Markets · 17 September

The first Fed hike since 2023 lands on the AI trade

The Federal Reserve raised its benchmark rate by a quarter point — the first increase in three years and the first under chair Kevin Warsh — after futures markets put the odds above 90 per cent and with traders already pricing three to four hikes by next summer. Blue chips took the hit while AI and semiconductor names held firm, an inversion of the usual script that reads as the buildout being treated as infrastructure rather than speculative duration. Huang used the week to tell Trump he would not allow an AI slowdown. The Fed's July minutes had already flagged that an AI disappointment "could lead to a significant repricing of stocks" — making the sector's reaction function a macro variable in its own right. The test is whether that tolerance survives a second hike.

Source: MarketplacePrimary: CNBC

Calendar

What to Watch

  • Today

    Huawei Connect, Shanghai — day two

    The conference runs 17–19 September. Watch for detail on the 960 super-node scale question: the 4,096-NPU system announced this week is well short of the 15,488-chip Atlas 960 SuperPoD Huawei had previously described.

  • 24 Sep

    Trump–Xi summit and state dinner, Washington

    AI governance is the principal agenda item. Altman, Huang and Cook are expected at the table. Whether Anthropic's Dario Amodei is invited decides whether this is an industry conversation or a chip conversation.

  • 29 Sep

    OpenAI DevDay, Fort Mason, San Francisco

    The keynote is livestreamed free. Astra shipped on 3 September; DevDay is where the developer surface for it gets defined, and where the deprecation of the last legacy GPT-3.5 endpoints on 28 September becomes concrete for anyone still pinned to them.

  • 1 Oct

    Apple v OpenAI hearing

    Apple's trade secrets suit against OpenAI and two former Apple employees reaches a hearing on OpenAI's motion to dismiss. Separately, Musk's X Corp dropped its antitrust claims against Apple on 14 September but kept the case alive against OpenAI.

  • Mid-October

    Anthropic's Nasdaq listing

    Marketing for the initial public offering is expected to begin mid-October at the earliest, with the listing targeted for the days before the US midterm elections in November — the first real price on a frontier lab.

  • 17 Nov

    Nvidia Q3 results

    The clearest read on whether the infrastructure buildout is still compounding, and on how much of the Vera Rubin performance story the company chooses to tell publicly.

  • 2 Dec

    International AI Summit, Brussels

    Policymakers and industry on global governance and access, with the EU AI Act's high-risk obligations legally in force since 2 August pending the Digital Omnibus delay.

  • Q1 2027

    Ascend 960DT ships

    Huawei's brought-forward date is the first credible test of whether Chinese silicon can substitute for restricted Nvidia parts at frontier scale — and whether the super-node architecture holds together at the advertised size.