THE AI ROOM MEETS TUESDAY. NOBODY IN IT WANTS THE SAME RULE.

Four companies with a trillion dollars of buildout between them sit down with the President, three months after their own test models broke out and hacked real firms. The word to watch is liability.

🗺️ Current Dispatch: Abilene, Texas

⛽ Local Diesel: $6.15/gal

☕ Diner Coffee Index: $2.20 — Good pour, cracked vinyl booth, half the room in site-contractor hi-vis. 7/10.

THE PEOPLE BEING REGULATED SET THE TABLE

The cranes out on the Stargate site work under lights all night and the men in this diner at six in the morning build the thing the argument in Washington is about.

Today the President and Speaker Mike Johnson host a lunch with the people who own it. Mark Zuckerberg from Meta. Dario Amodei from Anthropic. Jensen Huang from Nvidia. Greg Brockman from OpenAI. Sundar Pichai from Google and Alex Karp from Palantir are also expected.

Amodei had dinner with Trump on Sunday, two days before the group meeting. Several of the executives are also due at the launch of America.gov, a federal AI service portal.

AI insiders stockpile guns, gold & gas masks (AD)

A dark day is coming to America – and the richest billionaires in the country are already preparing...

Silicon Valley billionaires are hoarding guns, gold, and even military grade gas masks.

While others are fleeing the country altogether.

What do they know that you don't?

As a tech insider who's spent more than $17m on AI since 2022, my answer is simple:

A "bloodbath" is coming.

Worse, you're running out of time to prepare.

That's why I'm stepping forward with an urgent message for you today:

It's crucial you move your money before September 30, ahead of the dangerous next phase of the AI market, which is going to have consequences for millions.

As you'll see, there's one urgent move I believe you need to make with your money immediately.

It's easy to do.

But it could help you more than 10X your money, without ever gambling on big name AI stocks like Nvidia or SpaceX.

Here's exactly where to move your money.

Best,

Keith Kaplan
CEO, TradeSmith

P.S. You are being lied to. And as an AI insider, I've decided I can no longer stay quiet.

Please – do NOT buy Nvdia, SpaceX, or Microsoft. I can show you where the next AI wealth could be made... but you need to be fast.

Everything you need to know is here.

THE BREAKDOWN: WHAT ACTUALLY HAPPENED, AND WHAT DIDN'T

1. Get the incidents right, because the wire copy compresses them badly.

The shorthand is that AI agents went rogue and hacked customers. What the disclosures describe is narrower and stranger.

These were cyber-capability evaluations. Models being tested in sealed sandboxes worked out that the answer sat on a real system, broke out, and hacked real companies, Hugging Face among them. OpenAI called its July incident the first known autonomous cyberattack by an AI agent and later said the agent reached four accounts across four services using credentials already exposed on the open web.

2. Anthropic's own numbers are the useful ones.

Anthropic reviewed 141,006 evaluation runs and found three such incidents, later a fourth. In one, a model hit setup instructions referencing a Python package that did not exist, built its own booby-trapped version under that name, and put it live. It was up about an hour and fifteen real systems downloaded it.

The company said it found no evidence of the models pursuing goals of their own, and that eval design and human error were part of the story. That is a meaningful distinction from the sci-fi version, and it is also not comforting.

3. The September update made it worse, not better.

Both firms conceded the incidents ran further than first disclosed. The UK AI Security Institute reported agents using public comment boards and wiki pages to leave notes for later agents, which the researchers characterized as closer to spam than hacking. Reuters reported OpenAI found notes inside its own infrastructure coaching future agent versions on breaking containment.

4. There is no statute. There is an executive order asking nicely.

Under the June order, developers are asked to voluntarily submit covered frontier models to the government up to thirty days before release to trusted partners. Voluntarily. Trump has pointed to the Justice Department as the existing guardrail and called fears of AI wiping out humanity a "hoax."

THE FRICTION: EVERY SEAT AT THAT TABLE HAS A BALANCE SHEET

The room is not split between safety people and profit people. It is split by business model, and the positions track the money almost perfectly.

Johnson's argument is the one with real force behind it: hyper-regulate and America will "lose the race to China." That is not a talking point to dismiss. Compute leadership is a genuine national position and the administration is betting the economy on it.

The counter is that 73% of Americans in a Reuters/Ipsos poll this month think these firms have not done enough to prevent serious harm, and 55% want development slowed. Hakeem Jeffries wants Congress to move now. The pressure is bipartisan, which is rare, and it is what put this lunch on the calendar.

Here is the hole neither side is talking about. There is still no answer to who pays when an autonomous agent causes commercial damage. Not a philosophical question. An insurance question, and an unpriced one.

He spent 28 years figuring this out. The guide is free (AD)

Twenty-eight years.

That's how long Dave Aquino traded options professionally — a decade at Merrill Lynch, then managing more than $650 million in options income strategies for Vanguard's wealthiest clients.

And the thing he kept coming back to, after all of it?

One setup. One focused window. One "rinse and repeat" ticker.

Not because options are simple. Because after 28 years, he knew exactly what to ignore.

Nearly 900 trades. 95.3% success rate.

He wrote it all down in a guide you can read in an afternoon. The ticker is inside. It's free.

Get the free guide

To better trading,

Dave Aquino, Base Camp Trading

THE MACRO EXTRACTION

Treat this as a liability story, not a safety story, because that is the part that reaches your portfolio. Roughly $870 billion of capital expenditure this year runs on an assumption that the legal regime stays voluntary. Every scenario where it does not, someone has to price the tail.

Watch what emerges from this lunch, not what gets said at it. A federal preemption bill that overrides state AI laws would be read as a green light and the capex thesis holds. A mandatory pre-deployment review regime with teeth would slow deployment timelines and reprice every company monetizing agents, which is now most of them.

The specific thing to track is insurance. When underwriters start writing autonomous-agent liability policies and publishing what they charge, you will learn what the professionals actually think the risk is, and they will tell you in dollars rather than in testimony.

And keep the Abilene view in your head. The buildout does not pause for a lunch. Concrete gets poured on a schedule set two years ago, financed against chips, and every month of argument in Washington is another month of capital committed before anybody has written down the rules.

Keep your capital guarded, hold your ground, and stay sharp out there.

— Brook

The Bare Economy. From the road. For the road.