Guardians of the Frontier

Where Amodei, Altman, and Musk all agree... Regulation could be coming – or not... 'Don't kill the Golden Goose!'... What it all costs... The Federal Reserve meets this week... The market expects a rate hike...

The 'Heads of AI' agreed to 'slow down' this weekend...

On Saturday, Anthropic co-founder and CEO Dario Amodei published a post on his personal website titled "We Must Pace the Frontier." His industry peers, Sam Altman and Elon Musk, made their own public statements in agreement.

Here's the line from Amodei's weekend missive that got a lot of people's attention. Amodei himself wrote it in boldface...

We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.

Amodei runs one of the world's most powerful AI-development firms. His fear is "recursive self-improvement." That means AI systems themselves creating better AI systems faster than any human can understand or control.

He referenced the recent "OpenAI-Hugging Face incident." In case you missed it, a group of AI agents got online, coordinated with each other, and launched cyberattacks – all to cheat on an evaluation.

Amodei said he worries that within six to 12 months, a "swarm" of such models could be capable of taking over the entire Internet, "potentially causing hundreds of billions of dollars in damage."

He proposed three steps for "pacing the frontier," in descending order of how likely they are to happen...

1. Embedded third-party evaluators – outside inspectors with desks, badges, employee-level access, and the right to publish what they find without the company editing it. (To me (Corey McLaughlin), that sounds like proactively opening the door to government regulation.)

2. A common set of safety standards among the frontier labs headquartered in democracies, which would need approval from federal antitrust regulators. (Same.)

3. An agreement with authoritarian governments – China above all – which Amodei himself conceded is the "toughest dilemma" in the plan. (Good luck with that one... China will dominate AI development if it can.)

Amodei committed Anthropic to at least the first step.

This essay did not arrive out of nowhere...

Earlier that week, as my colleague Nick Koziol reported in last Wednesday's Digest, former Anthropic and OpenAI researcher Jacob Coxon had just publicly resigned... Coxon wrote that "neither company is acting responsibly" and that the industry is "racing straight to self-improving superintelligence and gambling with our lives."

The real humdinger: Anthropic's own alignment science lead, Evan Hubinger, posted this on the social platform X...

We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade.

That is a man in charge of making sure the models "behave" putting a number on it in public, and with an exclamation point no less. It's like he's excited about it... making everyone wonder what the heck these AI labs are really doing.

Amodei's Saturday essay reads like something of a response to the groundswell from X posts (which I can't really believe I'm writing) that generated tens of millions of impressions.

Peer endorsements came quick this weekend, as if they were arranged on some private messaging group...

Altman, OpenAI's CEO, said Saturday he agrees "that we need to pace the frontier"... and that OpenAI "will do the same" with outside evaluators.

Then Elon Musk, who has been public about putting constraints on AI development for years, shared Amodei's post on X with three words: "Dario is right."

Meanwhile, Google DeepMind CEO Demis Hassabis backed the direction while saying the details need work.

So, you might expect that some or any regulation in AI will be coming down the pike – or not...

We've seen this time and time again from emerging tech or industries... If they sense the winds are shifting again them, they make proactive PR campaigns.

They'd rather cooperate with the government (or pretend to) than face blowback from either regulators or the public.

Though it doesn't sound like they have much to worry about from the U.S. government – right now anyway. Here's President Donald Trump today on his Truth Social platform, warning everyone not to mess with the industry behind a big chunk of America's recent GDP growth...

On that point, none of the AI CEOs said anything suggesting their spending plans would change. Nobody announced a delayed model release, a slower cadence, a capability cap, or a single dollar less of capital spending.

Follow the money, not the essay...

Consider what Anthropic has committed to.

The company has now pledged to buy $517 billion worth of chips and data-center infrastructure covering 14.8 gigawatts of capacity – with four new deals struck since July alone. Not long ago, its disclosed commitments ran to "only" $180 billion through 2029.

So while its CEO talks up guarding the frontier, Anthropic has just about tripled what it's planning to owe chipmakers and data-center landlords.

And Anthropic is about to ask the public to help pay for it. Reuters reported Friday that the company is seeking about $100 billion at roughly a $2 trillion valuation. Nvidia is in talks to lead the way by investing up to $10 billion. Marketing should begin in mid-October.

The company is telling investors it was profitable for a second straight quarter. Its reported revenue has gone from $9 billion to $65 billion in seven months, with gross margins above 80%. Still, those numbers carry a big asterisk.

According to the Financial Times, those 80% margins are "before accounting for revenue shared with distribution partners and the cost of training its models."

Large details like this are why we remain concerned about the buildup of an AI bubble that could burst when enough companies fail to fulfill their big promises. And as our colleague Whitney Tilson wrote in his daily e-letter today...

If it bursts, it's hard to tell how widespread the fallout could be.

Meanwhile, OpenAI is planning to spend around $600 billion by 2030. And while its IPO has been delayed, it still wants investors' money, too.

Since Anthropic and OpenAI are still privately owned, they don't have to share as much financial data as a publicly traded company.

Public companies' filings tell a similar story on AI spending...

Alphabet (GOOGL) projects up to $205 billion in capital expenditures this year. Microsoft (MSFT) expects to spend about $190 billion. Meta Platforms (META) is looking at more than $145 billion – a bill Mark Zuckerberg has said out loud is being funded in part by cutting 8,000 jobs.

Add Amazon (AMZN), and those four companies alone are on pace to spend some $750 billion in 2026. That's up more than 80% from last year's record $410 billion.

None of that changed over the weekend, either.

We wrote in last Tuesday's Digest about the hyperscalers crowding Uncle Sam out of his own bond market... and Credit Opportunities editor Mike DiBiase's observation that these companies are essentially out of free cash flow and have to borrow to keep building.

Goldman Sachs sees hyperscalers issuing $250 billion of debt this year and $400 billion in 2027. A real slowdown in AI spending would show up in this trend first. In the past three days, it has not.

What the weekend produced was something of a public-relations campaign and a promise to rein in development of AI models. That's probably worth doing, but it doesn't change these companies' plans to spend, spend, spend.

Still, the market reacted with some concern to all this today. The semiconductor sector was down nearly 6%, and many individual AI names like Corning (GLW), Coherent (COHR), and Lumentum (LITE) were off much lower. The major U.S. indexes finished slightly lower.

Meanwhile, the bill for everything else keeps climbing...

The 10-year Treasury yield sits at 4.99% as of this writing, up from 4.80% when we wrote about it last week and essentially at 5%. That's a level it has visited only briefly since the financial crisis... when yields spiked amid inflation concerns in October 2023.

Higher Treasury yields mean slumping demand for U.S. government debt.

Meanwhile, in the Middle East, Iranian-backed Houthi militants seized more islands in the southern Red Sea and shut down a major Saudi oil pipeline. Oil prices jumped in response. The international benchmark Brent crude's November contract was up more than 1% to $106 a barrel. And the U.S. benchmark West Texas Intermediate's October contract rose to near $102.

The national average for gasoline is $4.32 a gallon, up 17 cents in a week. Diesel is $6.23.

This brings us to the big economic event of the week: the Federal Reserve's meeting, led by Kevin Warsh, which runs tomorrow and Wednesday.

Last week, we noted that roughly 58% of federal-funds futures traders were pricing in a 25-basis-point increase at the meeting.

As of this afternoon, fed-funds traders' expectations for a hike have jumped to 92%. They're convinced.

All the data has pointed to persistent "sticky" inflation in recent weeks and months, and the market is confident that the Fed will fight it with higher interest rates.

The AI executives spent their weekend saying they'll develop their AI tools more carefully. And as bond investors sort through all this AI spending, the U.S. national debt, and inflation, they're selling Treasurys.

As AI spending continues, it's tempting to buy AI stocks...

But Whitney Tilson warns that it's a risky move.

Whitney – a former hedge-fund manager and the editor of several Stansberry Research newsletters – has been saying for weeks that 2026 is a "tentpole year." That's what he calls a handful of years, like 1999, that investors remember by feeling rather than by date.

Right now, he says it feels like he has gone back in time to '99 – and the build-up to the dot-com bubble.

Whitney isn't predicting that the AI build-out will stop. He just remembers what happened to the money that chased the peak of the last big tech bubble... and the stocks you can own in a market that is headed toward the same fate...

In a presentation he debuted last Thursday, Whitney laid out the screen he calls the "Greenwich Test" – and named a list of companies that meet it. They're not the AI names you see in the headlines... Instead, they include an HVAC contractor, a medical-device maker, a gold miner, an optical-components manufacturer, and a steelmaker.

All have delivered triple-digit gains in recent years.

If you aren't familiar with more obscure stocks like these, it's time to get acquainted, as the AI boom risks turning into a euphoric bubble...

In a near-decadelong back test, Whitney found that the Greenwich Test's stocks would have turned $100,000 into $1.1 million since 2017. In the past 12 months alone, Whitney says, they could have doubled your money or more on 46 separate occasions.

In his presentation, Whitney shares more about why today's market reminds him of 1999, along with more about this strategy.

He also names his No. 1 stock to buy today – free of charge – and one big-name stock he says to sell immediately.

Click here to watch the replay before it comes down.

New 52-week highs (as of 9/11/26): Alpha Architect 1-3 Month Box Fund (BOXX), Chord Energy (CHRD), Chevron (CVX), iMGP DBi Managed Futures Strategy Fund (DBMF), iShares MSCI Japan Index Fund (EWJ), Franklin FTSE Japan Fund (FLJP), Global X MSCI Greece Fund (GREK), Hewlett Packard Enterprise (HPE), Plains All American Pipeline (PAA), and Valero Energy (VLO).

In today's mailbag, a question about our annual Stansberry Research conference, which is coming up in just a few weeks in Las Vegas... Do you have a comment or question? As always, e-mail us at feedback@stansberryresearch.com.

"Can I access past years' annual conferences?" – Stansberry Alliance member Andrew J.

Corey McLaughlin comment: Each year, we provide video replays of our Stansberry Conference speakers through the end of that year. We also have a full archive of our Alliance Day presentations, which you have access to as one of our Alliance members. You can find that archive here.

This reminds us... Our 2026 Stansberry Conference is coming right up on September 28 through 30.

All our subscribers can access the first two days of this year's conference with an in-person ticket or a Livestream Pass. This year's in-person tickets have sold out, but we still have the Livestream Pass available. Click here for more information.

2026 is shaping up to be our best event yet, with incredible speakers like famed actor Henry Winkler (aka The Fonz!), highly sought-after tech expert Dan Ives, biomedical researcher Dr. David Agus, bestselling authors, CEOs, entrepreneurs and innovators, and more...

You'll also hear top ideas and stock recommendations from Porter Stansberry, Dr. David "Doc" Eifrig, Eric Wade, Dan Ferris, Greg Diamond, Marc Chaikin, Joel Litman, and others.

And Alliance members, if you're not going to be with us in Vegas, your membership includes streaming access to Alliance Day. That's the third and final day of the conference. As the event gets closer, we'll tell you how to log in to watch your exclusive Alliance Day livestream. To access the first two days, sign up for the conference's Livestream Pass.

All the best,

Corey McLaughlin
Baltimore, Maryland
September 14, 2026