The Download

Morning Brew Inc.

TL;DR: Today, Nvidia launched an alliance to build open AI cyberdefense tools… and the biggest names missing from it are Anthropic and OpenAI. Their absence is a sign of a widening Silicon Valley rift: Should the most powerful AI models be open to download and tweak, or locked behind a company’s API?

What happened: The Nvidia-led Open Secure AI Alliance—founding members include Microsoft, SpaceX, IBM, and more—follows a letter Nvidia and over 50 other firms signed last Friday asking the government not to place “premature restrictions” on open models. Anthropic and OpenAI did not join today’s group of avengers, though OpenAI did sign Friday’s letter after it was published.

All this public posturing comes after OpenAI’s models autonomously hacked AI platform Hugging Face earlier this month. To investigate, the site turned to an open Chinese model from Z.​ai, saying safety guardrails limited how useful frontier models like Anthropic’s Fable were for the job.

The Silicon Valley split: The pro-open-weight camp argues that self-hostable, modifiable models bolster AI safety, since your defenses don’t hinge on someone else’s locked-down system. Open models also boost innovation and widen access to AI—handy for your bottom line if you’re in the business of selling AI chips, like Nvidia. (It’s worth noting Nvidia is financially entangled with closed-model labs, too.)

The other side sees danger. AI is now too powerful to offer without a leash, especially as recent Chinese open model releases show how thin the US-China capability gap has gotten. Competition from open models also puts pressure on what proprietary labs are able to charge for their products.

All eyes on Washington: White House officials (here are which ones, exactly) are mulling how to crack down on distillation without stifling innovation. Insiders say regulation rather than a blanket ban is currently more likely, per the New York Times. Anthropic and OpenAI have both privately lobbied for tighter restrictions; Altman meets administration officials this week to preview unreleased models and will likely weigh in on the open-weight debate.

Bottom line: The open-versus-closed AI fight is now an active lobbying war. Expect the debate to heat up this week as Amazon, Meta, and Microsoft report earnings. —WK

      

Sponsored By State Street Investment Management

Looking for an easy way to capitalize on the latest trends? Invest in the AI and tech innovators transforming America’s future—like Nvidia, Tesla, and more. SPY offers you exposure to the world’s largest economy and many of its most recognized brand names in a single trade. Capture the growth potential of the S&P 500 with SPY.

The Zeitbyte

MAXimum effort

Imaxxing, Adobe Stock

Sure, Odysseus spent a decade fighting sea monsters and cannibalistic giants to get home, but he’s got nothing on the people spending big money and making very long treks to see Christopher Nolan’s latest film in 70mm IMAX.

Only 25 theaters in the US can screen the blockbuster in this special format, with seats in such high demand that resale tickets on eBay are going for over $1,000. To help with this crisis of cinema, one San Francisco developer launched (what else) IMAXXING, an automated service that monitors theaters and emails you when good seats open up. Its homepage features an illustration of infamous looksmaxxer Clavicular in ancient Greek armor and commands you to “get frame mogged” and “start retina-smashing today” by signing up. But if this bot doesn’t do it for you, you can always wait to watch The Odyssey on a tiny screen in a middle seat in economy, a journey even Odysseus himself might have found grueling. —WK

Chaos Brewing Meter: ☕☕/5

• A rumored $250 billion Nvidia-OpenAI infrastructure deal is raising concerns about circular financing.
• If you yearn for the days of a fax machine or for a LinkedIn feed that isn’t littered with AI-slop adverbs like “genuinely” or “quietly,” you’re not alone. Two-thirds of white-collar workers are nostalgic for their pre-AI workplace.
• Google search results reportedly surfaced a trove of Claude users’ chats and artifacts, revealing things like SSNs and medical records. This is the common sharing setting that caused it.
• Here’s how long it took OpenAI to notice that massive Hugging Face breach.
• Apparently, even a low-stakes leak was enough to send DeepSeek’s “normally boring” CEO into a tailspin.
• If you’ve ever wondered what a North Korean cellphone looks like or what you can do on it, wonder no longer.
• Apple’s smart glasses debut got pushed back to June 2027, partially because of privacy concerns.

This time last week...

Readers’ most-clicked story was about when Altman invited author Dave Eggers to speak at the OpenAI office—the result was probably not what Altman expected.