Jensen Huang apology and investigation & Goldman Sachs restricts AI coding tools - Hacker News (Nov 9, 2026)

The CEO of one of the most valuable companies on the planet just sent an apology to his family, his staff, and his shareholders — and tucked inside it was a mention of a national security investigation. We'll get to that in a moment. Welcome to The Automated Daily, hacker news edition. The podcast created by generative AI. I'm TrendTeller, and today is November 9th, 2026. On the list: Wall Street pulling back on who gets to use AI coding agents, a massive language model squeezed onto an ordinary computer, a connected-car hack that started with nothing but a license plate, Outlook quietly bouncing your email, and a few thoughtful reads on software and writing. Let's get into it.

Jensen Huang apology and investigation

First, NVIDIA. The company published an internal message from Jensen Huang, dated November 8th, in which he apologizes to his wife, children, employees, board, partners, and shareholders. Huang says he was unfaithful, takes full responsibility for the harm caused, and asks for privacy for his family. The more consequential line for the business world is this: NVIDIA's board has opened an independent investigation into national security concerns mentioned in news reports, and Huang says he'll cooperate fully. Personal matters aside, a board-level probe touching national security at the company powering most of the world's AI infrastructure is something investors and regulators will be watching closely.

Goldman Sachs restricts AI coding tools

Staying with AI and big institutions, Goldman Sachs has restricted Anthropic's Claude Code and Microsoft's GitHub coding tools to its engineers, cutting off most traders and bankers. That's a notable reversal, given the bank's CIO had recently praised non-technical staff for building useful AI tools. Goldman's reasoning is that code-writing agents should stay with people trained to review and govern what they produce, since a trader's homemade script can touch client data and market-sensitive workflows. Goldman had already limited Claude in Hong Kong earlier this year, but this policy is broader because it's based on role, not location. And the relationship with Anthropic isn't cooling overall — Goldman still works with them on back-office projects and is a lead underwriter on Anthropic's planned IPO. It's a clear signal of how banks are trying to balance AI enthusiasm with compliance reality.

Giant AI models on CPUs

On the opposite end of the spectrum, here's a project trying to put huge AI models in everyone's hands. A GitHub repo called llama-modchip offers patches and tools for llama.cpp that run large mixture-of-experts models entirely in system memory on a regular CPU. The headline claim: a 284-billion-parameter DeepSeek model fitting on a machine with just 64 gigabytes of RAM, thanks to aggressive compression and careful memory tuning. It also targets smaller models on 16 to 32 gigabyte systems and adds features aimed at snappy, interactive agent use. If the performance holds up, it's another step toward high-end local AI without needing a GPU at all.

Connected-car hack from license plate

Now to security. Researchers Shubs and Sam Curry describe how they found a chain of flaws in a major automaker's connected-car platform that let them control vehicles remotely. It began with reverse-engineering the mobile app, which led to exposed admin tooling and hardcoded credentials sitting in JavaScript files. From there, weak identity endpoints, a firewall bypass, and a login flow that skipped multi-factor authentication let them take over accounts and send real commands, like unlocking doors, in the automaker's Australian environment. The starting point? Just a license plate. No single bug was catastrophic on its own — the lesson is how small misconfigurations stack into full remote access.

Outlook quietly enforces DMARC records

Email admins, this one's for you. Microsoft's consumer Outlook service — covering outlook.com, hotmail, live, and msn addresses — started rejecting some messages from domains without a DMARC record on November 5th, with no public announcement. The tricky part is that only some messages bounce, so a domain can look fine while mail quietly goes missing. The error mentions missing authentication records generally, but many affected senders already have SPF or DKIM and simply lack DMARC. The fix is easy: publish a basic DMARC record in monitoring mode, which won't change how other providers treat your mail. The bigger picture is that DMARC has moved from best practice to baseline.

Declarative Dragging wins UIST award

Over to interface research. A project called Declarative Dragging won Best Paper at UIST 2026. Instead of hand-coding how a dragged object behaves, developers describe the states it's allowed to end up in, and the system previews those possibilities to figure out where things can land and how they should animate. It ships as a library called Dragology for web SVG interfaces, with demos in math and diagramming. The appeal is making rich, tactile drag interactions far less painful to build, especially for education and design tools.

BenchUp open-source woodworking modeler

For the makers out there, BenchUp is a free, open-source 3D modeling app aimed at woodworking and furniture — think shelves, tables, and workbenches planned to exact measurements before you cut anything. It runs in the browser or as a desktop app, stores everything locally with no account required, and can import SketchUp geometry. It's a refreshingly simple option for DIY builders who don't want a subscription attached to their sawdust.

Harvest price hike after acquisition

Speaking of subscriptions, a longtime user of Harvest, the time-tracking and invoicing tool popular with freelancers, says it's been effectively ruined since Bending Spoons acquired it in 2025. The author's monthly bill reportedly jumped from about 13 dollars to nearly 134 after being pushed onto a new enterprise plan, with little visible change in service. A cheaper tier now takes an extra cut of invoiced revenue on top of card fees, and support emails are apparently answered by a language model. It's a familiar story of acquired software getting pricier and less personal — and small businesses feel it most.

Writing as a thinking tool

Finally, a reflective essay arguing that writing isn't just for recording thoughts — it's how we create them. The author draws on Manuel Blum's idea that writing acts like external memory, multiplying our problem-solving ability, and describes how drafting forces deeper understanding of papers and ideas. The warning is that LLMs make it tempting to skip that step and produce polished text without real comprehension. The advice is practical: draft messily, edit ruthlessly, keep it simple, write for yourself first, and make it a habit. Writing slows you down just enough to think — and that may matter more now than ever.

That's the rundown for November 9th, 2026. From boardroom investigations to bounced emails, it was a day of reminders that trust — in leaders, systems, and tools — is easy to lose and hard to rebuild. Links to all the stories are in the episode notes. I'm TrendTeller, thanks for listening to The Automated Daily, hacker news edition, and I'll catch you tomorrow.

Jensen Huang apology and investigation & Goldman Sachs restricts AI coding tools - Hacker News (Nov 9, 2026)

Our Sponsors

Today's Hacker News Topics