Read Online | Sign Up | Advertise | | | Good morning, C, and welcome to our 4,617 new readers. Nvidia’s chips are at the center of the U.S.-China AI race, and Washington has spent years trying to keep the best ones out of Beijing’s hands. | Now, a New York Times report says U.S. intelligence found that an alleged Chinese asset had built close ties to Nvidia CEO Jensen Huang… and the White House chose not to act. | | In today’s AI rundown: | Report: Nvidia CEO targeted by alleged Chinese spy Rowan’s Corner: ‘Sent from my agent’ Make ElevenLabs voiceovers sound more natural Anthropic’s AI now monitors its own conversations
|
|
| | | | NVIDIA | | | The Rundown: The NYT just reported that U.S. officials found Alice Huang, an exec at Singapore-based data center firm Megaspeed, was allegedly passing intelligence to Beijing while keeping close ties with Jensen Huang, with Trump briefed about a year ago. | The details: | It’s unclear what Ms. Huang passed to Beijing or its value, but officials worried she could offer insight into Trump’s view on chip sales to China, and into Nvidia itself. Officials didn’t believe Jensen knew of her alleged role when Trump was briefed; he never got a defensive briefing, often given to U.S. citizens targeted by spies. Megaspeed, already under U.S. probe for steering Nvidia chips to China, bought at least $4.8B in Nvidia gear via a sister firm, and remains an Nvidia cloud partner. Nvidia called Jensen’s meetings with Ms. Huang routine, saying customers get ordinary business info and “zero information relevant to national security.” Megaspeed said outside counsel is investigating and it’s cooperating with the U.S., while China called the claims “groundless smears and slanders.”
| Why it matters: The report reads like a spy thriller, but nothing has been publicly proven, including whether Ms. Huang passed anything to China. The immediate fallout may hit the loophole allowing firms like Megaspeed to rent Nvidia-powered computing from Southeast Asia to Chinese AI labs — one Trump reopened but may now close. |
|
| | TOGETHER WITH OPENART | | | The Rundown: Need a launch video for your business? Simply paste your website link, add your brand assets, and define your target audience. OpenArt’s AI video agent will shape the story and messaging while handling the script, motion design, voiceover, and video editing. | With OpenArt, you can: | Turn your website, screenshots, or pitch deck into a launch video Create short promos or longer videos to showcase your business Refine scenes, voiceovers, or music by describing the changes Go beyond launches with product ads and UGC-style videos
| Create your launch video with OpenArt. |
|
| | ROWAN’S CORNER | | | Rowan: “Sent from my agent” is becoming the new “Sent from iPhone.” That’s a tweet my friend Greg Isenberg recently made that’s stirring up conversation. | Basically, “Sent from my iPhone” started as a way to buy forgiveness for typos and messy formatting when emailing someone from your iPhone. But when someone sees “Sent from my agent,” the real message they receive is: you weren’t worth my attention. | There’s real tension here (for me included). Agents are the biggest leverage available right now, and I run several tasks through them, saving me at least a few hours a week. But I also believe going the extra mile and writing something yourself shows the other human respect. | So here’s my line: | Anything that touches a person is me — social posts, messages, recruiting calls, live video. Anything that’s a task, an agent can do.
| I also have a growing thesis that real human relationships will only become more valuable as AI gets better. Anything that risks that trust is probably a bad trade. Personally, if I see “sent from my agent,” I’m deleting the email. | I’ve written before that AI intuition is the most important skill of this decade, knowing which tasks to hand to AI, and this is the perfect example. Use agents everywhere, except on people. | Are you using agents to respond to your emails, or do you think the human touch shows respect? Let me know in the poll at the bottom of the email. I’m curious! |
|
| | AI TRAINING | | | The Rundown: In this guide, you’ll add Audio Tags to a short Eleven v4 voiceover, compare it with the plain version, and create a short scene with two speakers. | Step-by-step: | Open ElevenLabs Text to Speech, choose one voice, and select Eleven v4. Generate the guide’s plain script and save it for comparison Paste the tagged version without changing the spoken words. Keep the same voice and settings so you can hear the difference Compare the opening, pace, and final line. If the delivery stays flat, make one cue more specific. Change one thing at a time If a tag gets read aloud, check the model and square brackets. Keep the direction short and place it before the words it should affect Open Dialogue mode, add two voices, and generate the guide’s exchange. Check the words, speakers, and timing before downloading the clips
| Pro tip: Save a plain version before adding tags. A more expressive read may still sound less natural. You can find our full cheat sheet, annotated scripts, and review checklist here. |
|
| | PRESENTED BY GLEAN | | | The Rundown: Business data spans across apps, warehouses, and BI platforms, while the context behind the numbers lives in documents, tickets, and conversations. Glean’s upcoming webinar helps you bring it all together, find trusted answers, and make data-driven decisions without SQL or a BI background. | In this webinar, you’ll learn how to: | Query business data in natural language Quickly build dashboards to track key metrics Connect metrics with enterprise context
| Register for the webinar and put AI to work. |
|
| | ANTHROPIC / AI SAFETY | |  | Image source: Anthropic |
| The Rundown: Anthropic introduced “self-assessment,” a technique where Claude reviews the chat after each reply and flags potential misuse, saying the system has been active since June and catching large-scale abuse that might otherwise be missed. | The details: | Claude gives each conversation a quick safety verdict (whether it’s harmful, what kind of harm, and how urgent), which helps prioritize cases for human review. On Claude Fable 5, it catches 78% of potentially harmful conversations flagged by a stronger classifier, and 94% of high-priority ones, at a 1% false-positive rate. It helped uncover 100K+ child-abuse flags from a single API customer, reported to authorities, plus an automated romance-scam operation run through Claude Code. Anthropic admits it’s weak as a lone blocker, cutting attack success in red-team tests only from 79% to 54%, though it made each successful attack cost ~4x more.
| Why it matters: After Anthropic’s own reports showed Claude being misused for cyberattacks and weapons work, letting the AI check and flag misuse in its own chats is a welcome layer. Claude brings everything it understands to each judgment, but it can also share the same blind spots, meaning the system isn’t foolproof (yet!). |
|
| | | | COMMUNITY AI WORKFLOW OF THE DAY | ▸ Michael used Claude Cowork to build a 3D home design app | Today’s workflow comes from reader Michael Saif: | “I built this web app for a friend who is building a house. It lets you see the finished version of every room in the planned home before anything is built. I described the idea to Claude Cowork, had a short back-and-forth answering its questions, and it built the app in four stages. You can upload one or more floor plans, select a room, and view it from several rotatable 3D angles. | You can then describe how you want the room to look when completed, such as ‘Show me with a dark-colored wood floor with blue walls,’ and ask for changes, like light-colored wood floors. You can also add furniture to see how the room will look when finished, save multiple designs for each room, and export any image as a PNG. The whole build took less than an hour from start to finish.” | See the workflow on our Community Workflow Hub → | | How do you use AI? Tell us for a chance to be featured in the newsletter. |
|
| | | 🧑💻 Jotform App Builder - Jotform plugin/connector lets you build apps from AI conversations with ChatGPT & Claude* 🤖 Le Chonk - Mistral’s 1T-parameter AI for agentic coding, now with open weights 🖼️ Brush - Canva’s new vision-language model that edits designs via text 💬 Agent Memory - Databricks’ persistent long-term managed memory for agents
| *Sponsored Listing |
|
| | | VapiCon 2026, Nov. 11–12 in SF — explore how teams from Intuit to the NHS run voice AI in production, with speakers from OpenAI, Vercel, and more. Register and get half off with code THERUNDOWN50* | CEO Sam Altman said on the Cheeky Pint podcast that OpenAI will point its robotics program toward data centers, and that self-replicating data centers, with AI-controlled robots running fabs and building solar panels, are now within grasp. | Cohere announced a Series E targeting up to $5B to scale its models and infra, reporting 165% year-over-year ARR growth and a 20-fold customer expansion to 300. | Anthropic is reportedly SpaceX’s largest single customer, accounting for 34.5% of its Q3 revenue via AI compute contracts and spending reaching $3.1B/month from Dec. | Meta debuted Style Sketch, an AI app that builds virtual wardrobes from Instagram or uploaded images, suggests outfit combinations, and generates virtual try-ons. | Harvey acquired Aida, an AI sales startup founded by Glean and Patreon executives, to help build firm-specific ontologies that make internal knowledge more useful to AI agents. | *Sponsored Listing |
|
| | | | That's it for today!Before you go we’d love to know what you thought of today's newsletter to help us improve The Rundown experience for you. | | See you soon, | Rowan, Zach, Shubham, Jennifer, and Nate — the humans behind The Rundown | |
|
| |
|
