I laughed at it.

I rarely tell this story. It involves me, a phone call, and the worst seven words I ever said out loud. 

But you should hear it. Because I’m about to watch a few million people make the same exact mistake. 

Here’s what happened. 

My friend had just sold his company—a deal I helped put together, to the same buyer that took my first company, Reset. Life was good for him. He had money, options, and time.

He called me up. "You should come out to Silicon Valley. I'm moving there. My old college roommate is starting a company. He wants to build rockets to send people to Mars."

I remember my exact words.

"That’s the stupidest idea I ever heard."

Why would anyone want to go to Mars?

Understand the moment. 

The dot-com crash had just vaporized trillions. Rockets meant NASA and only NASA. Mars meant science fiction. Despite Mars being beyond the point, every smart person I knew would have said the same thing I said.

That was exactly the problem. Every smart person agreed with me.

My friend’s roommate, of course, was Elon Musk.

The sting faded, but the lesson didn't. 

In 2012, while everyone was still skeptical, I published my case for how SpaceX would turn a profit. In 2015, I wrote how it would change the economics of space. And years before a single share traded, I predicted Elon would pull off the biggest IPO the world had ever seen.

I could have been in the building. I wasn’t. But here’s what I learned: When a plan sounds stupid, especially coming from Elon Musk, look closer. 

Electric cars sounded stupid—Tesla is worth $1.5 trillion. Landing a rocket on its tail sounded stupid—SpaceX owns 80% of everything America launches. Beaming internet from space? Stupid. Starlink covers the planet.

And there’s a reason you need to hear this now. 

Earlier this year, Elon quietly filed a document with the FCC. No announcement. No tweet. Nothing. From a guy who sells flamethrowers for fun, silence is a signal.

I've read the filing. What's inside is bigger than Tesla, SpaceX, and X—combined. It solves the single biggest problem in the $25 trillion AI race. And it does it in a place none of his rivals can follow him.

I’ve put all the details together in my free Millionaire Maker Masterclass. 

But, before you check it out, I'll tell you this. 

The moment you hear what Elon is actually planning, your first reaction will be word for word what mine was in 2001:

"That is the stupidest idea I ever heard."

Good. Now you know exactly what that means.

Tesla minted the Teslanaires. The SpaceX IPO minted thousands more—welders, machinists, cafeteria workers. This wave dwarfs them all. By my math, he’ll make over a million new millionaires. 

So now I've done something I've never done before. I’ve created a blueprint on EXACTLY how to play it. 

I put the entire story—the filing, the plan, the missing piece, the crucial suppliers, and the tickers—into our free Millionaire Maker Masterclass. 

You don't need experience in the markets. And the research speaks for itself.

One warning: this all kicks off September 25th. After that date, I expect the headlines to catch up… and the window with them.

Twenty-five years ago, I hung up the phone. This time, I’m going all-in. 

CHECK OUT THE FREE MASTERCLASS NOW

He reads dead people's mail for a living. That's his job description. It's called being a historian, and apparently there's a PhD for it.

He wrote a book called How to Get Rich in American History. If you love money and hate history, you’re still going to like it. 

I went in expecting Rockefeller stories. Instead he told me about George Washington's diarrhea.

Martha’s Money 

In his mid-twenties, Washington was broke and, to make matters worse, he had a gambling problem. 

Meanwhile, Martha Custis was the richest widow in Virginia and every man in the colony was lined up at her door. 

Washington was maybe fifth in line. Tall, handsome, some military bravado. But certainly not the prize-winning stud. 

He only had one shot to get with her. And the night before, he got sick. Bad. 

He pulled himself together, showed up, and apparently did great, because she wrote him a letter afterward that Joe describes as “the 1758 version of Netflix & Chill."

Every cool thing Washington did after that was funded by him wooing a woman while suffering diarrhea. No small achievement. 

Then there's Franklin. 

Early to Rise, Deeply in Debt

Franklin printed pamphlets about the evils of debt. But it was all a front. He'd borrowed to buy his press, fallen behind, and the printer was threatening to repossess it. 

So Franklin went to a family in Philadelphia and made an offer. I'll marry your daughter if the dowry covers my debts. 

They said they didn't have that kind of money. Franklin said fine, mortgage your house. They declined. 

He married someone else.

That's the pattern Joe kept finding across 300 years. 

The Whole Family Goes to Jail

Steven Girard, a one-eyed cabin boy who became the richest man in America, borrowed almost everything he had to send one ship out. 

If it sank, he went to debtors' prison. And in 1790 debtors' prison took your wife and kids too. 

The ship came back. He paid the debt. 

Then he spent the next forty years borrowing less and less until he was sitting on 75 cents of equity for every 25 cents of debt.

How to Get Rich in America 

You might think the point of the story is to get in a bunch of debt and swing for the fences. But no. Debt doesn't make you rich. 

Joe was emphatic about this. Opportunity makes you rich. Debt is how you reach in and grab as much of the opportunity as you can carry. 

The people we remember grabbed once, got lucky, and then spent decades putting the leverage down.

That's the crux of Joe’s book: Get rich by taking the risk. Stay rich by dialing it down.

He boiled 300 years down to five things. 

→ Solve other people's problems.

→ Take risks, because this is the least risky era in American history and we're the most risk-averse Americans ever.

→ Move to where the opportunity is.

→ Marry well, meaning someone who covers your dreams while you cover theirs.

→ And believe you can, because the Consumer Financial Protection Bureau found that a positive attitude plus a savings habit predicted financial wellness better than income or inheritance.

Optimists live in mansions. Pessimists explain why you can't.

I asked him how much is enough. He said $2 million and half of Twitter would call him an idiot. He retired on it. Then his wife kicked him out of the house for asking too many questions about Foucault.

Get rich. Then dial it down. Then find something to do with your hands.

Listen to the whole conversation. It’s great. Joe’s great. I just wish he’d written the book sooner. For everyone’s sake. 

Recap: Perps are bets on price, with leverage, and no expiration date. And they've become insanely popular. As in: $60 trillion in volume per year. 

But there’s a reason maybe you’ve never heard of them: Americans were locked out. 

But that’s coming to an end. 

The CFTC blessed them in May and Chicago's big exchange, the CME, is suing to stop it. 

So the question today is twofold: Does it matter enough to boost any stocks? And, if it does, who’s going to win the lion’s share? 

I have ideas. 

Why this fight matters

Crypto perps alone are worth a few billion a year to whoever sells them. For Robinhood, that’s worth maybe a fifth of the stock. Real money. Not life-changing.

BUT… 

Stock perps are a different animal. 

The day an S&P 500 perp starts pulling from the 60 million options contracts Americans trade every day, this stops being a crypto story and becomes the biggest retail leverage market in the galaxy. 

That's the version the CME is going nuts over. 

That’s the version Coinbase and Robinhood want to dominate. 

THAT’S the version you want to pay attention to. 

Coinbase: the kitchen

On paper, Coinbase already won.

It spent two years building everything. A licensed futures exchange. A licensed broker to feed it. Deribit, the biggest crypto options venue on earth, bought for a few billion. 

And no less… it's been selling perps to Americans since last summer: bitcoin and ether first, then an S&P 500 perp at 20x this month.

And yet, read the quarter. Revenue down 19%. A $359 million loss, the third in a row. Monthly users down a million. Assets on the platform down 42%.

Coinbase built the best kitchen in the business and the dining room is emptying out. 

Its smart move was to cook for other restaurants: a stablecoin that earns interest, a derivatives exchange that Interactive Brokers resells, rails that Hyperliquid runs on. 

Half its revenue now comes from that, and it's the half that held up. Great kitchen. Fewer diners.

He can't open a Schwab account. He doesn't have a US broker. His market is closed when ours is open. 

Five years ago, that was the end of the story. Period. Full stop. 

Times have changed. 

These days, our guy in Jakarta can take a few hundred dollars of stablecoin and buy 20x Tesla at 3 a.m. No broker. No passport. No Wall Street in the loop.

And get this… 

Last quarter, a third of all trading on decentralized crypto exchanges were tied to stocks and other real-world assets. 

And it all has to do with a product called a perpetual future. Or, in short, a perp. 

Bets Without Bedtimes

Strip away the jargon and a perp is a bet on price with a running meter. 

You put up $100. You control $2,000. You bet the price goes up, or you bet it goes down. There's no expiration date, so you hold it for a minute or a year. 

The meter is the funding rate: a small fee that ticks between the people betting up and the people betting down, so the bet stays glued to the real price.

That's it. That's the whole product. And it doesn't care what the asset is. Bitcoin, oil, the S&P, Tesla.

The wild part: Seventy percent of all trading on crypto exchanges is perps. Last year, sixty trillion dollars' worth changed hands.

For scale, that's about a fourth of everything traded on every US stock exchange last year. One product, mostly illegal here, doing a fourth of Wall Street's volume in the shadows.

And this is the part that everyone got wrong. 

Cart Before the Horse, On Purpose 

Everyone expected tokenized stocks first, then the bets would be built on top of them. It went the other way. Perps came first. 

And the reason is clear:

A tokenized stock needs a regulatory framework. It needs someone to hold real shares, a custodian, an issuer, and an answer for the SEC. All a perp needs is a price feed. 

So the bet arrived before the thing being bet on. 

That's why our Jakarta trader was betting on Tesla long before he could buy the tokenized stock. 

And here’s why Americans should pay attention now. Until this spring, Americans haven’t been able to touch perps. 

The product lived offshore. That's how Binance became the biggest exchange in the world. That's how FTX got big enough to fall. 

The world got a 24-hour, leveraged, no-paperwork window into US stocks and crypto, and Americans were the only people standing outside it.

And it’s not like there’s no demand. Americans have been buying the clumsy version of this bet for a decade: options with a clock, leveraged ETFs that rot, margin with rent.

A perp drops the clock and the rot. The rent stays, but it floats, and half the time the other guy pays it. 

And now they’re coming home.

Panic at the CME 

On May 29, the CFTC approved a bitcoin perp on Kalshi—the prediction market. Same day, it cleared Coinbase to offer them. 

Coinbase now has 20x leverage on the S&P 500 for American customers. Robinhood bought a licensed exchange to launch its own. 

Then, in June, the Chicago Mercantile Exchange sued the government to stop it, calling it "textbook competitive injury." (Translation: they’ve seen the numbers. They don’t like what they see.)

But it’s too late. And it’s made up of two doors. 

Door one: the world buying America. Already open. Already big. Run by crypto exchanges nobody in Washington licensed.

Door two: Americans buying perps. Opening now. Fought over by Coinbase, Robinhood, Kalshi, and an incumbent with lawyers.

What they’re fighting over: a fee gets collected every time one of those ninety trillion dollars changes hands, at both doors. 

Somebody will stand at each one. The companies that get there are going to be big. The ones that don't are suing.

That's a war. 

It has contenders, a scoreboard, and dates that will decide it. Tomorrow, I’ll name the winners. 

Stay tuned.

And then… it face-planted into a padded blue wall and was carried off on a stretcher.

Another one cleared 2.88 meters in the standing high jump, past a human record that had stood since 1993. Others tripped, fell, broke apart. Some even caught fire. 

Twelve months after the first games, the humanoids are blowing past benchmarks. Some of them set by humans.

You might have questions. I’ll try to answer them. Then show you the ONLY place we’re looking. (Hint: Elon.) 

China’s NASCAR

The Humanoid Games are back for their second go and everything is bigger. Events went from 26 to 51, teams from 280 to 666, and 2,056 robots registered. 

But it’s more than just “robot Olympics.” 

Twenty-one events are robot worker scenarios: emergency response, hotel service, plugging in a USB port. 

That last one, as you might imagine, is MUCH harder than running on a track, and it's the one that matters.

So What? 

Robots have raced before. On wheels. Nobody cared. We solved the car a century ago.

But legs and hands are different. 

Walking upright was a big deal for humans. Hands are why we built tools. Add a brain that improvises and you have the recipe for a species. 

Watch a machine do all three, fall, and get back up alone, and something in the back of the skull registers a category change.

What Everyone Missed

The biggest story got almost zero coverage. 

The tennis robots run on a tiny computer doing 275 trillion operations per second. An unimaginable feat only 20 years ago. 

For perspective, the first computer to hit one trillion was in 1997. It filled a room bigger than a tennis court and cost $55 million. 

The tennis robot carries 275 of them in its chest, on a battery. 

It has to. 

A serve gives it a tenth of a second to see, decide and swing. The cloud can't answer that fast.

This is what people mean by “Edge AI,” and it’s nothing new. It’s been in your phone since 2017. But advanced Edge AI with legs has been two years away ever since. On Saturday the two years ran out, in 9.39 seconds.

You Can See It

Here’s why the games are so powerful. 

Most frontier tech is invisible. A model goes from 73% to 81% and you feel nothing.

Beijing made it visible. 

Two thousand machines running, fighting, playing football, in front of a paying crowd. 

It was the best marketing a government ever bought, and 641 of 666 teams were Chinese. In one weekend a large chunk of the planet decided China leads the frontier.

Maybe that’s wrong. But perception moves capital. Unitree IPO'd in Shanghai this week and went vertical.

Optimus Stayed Home

Sixteen countries sent machines to Beijing. But the most famous humanoid on Earth—Optimus—stayed home. 

It always does.

But in the US, the press is starting to fixate on one story: Tesla marrying SpaceX. A merger. 

Dan Ives just put the odds at above 80%. And however big you’re thinking this could be… it’s bigger.

Rockets, cars, batteries, chips, AI, and a humanoid. One roof. Nobody has ever owned that. And nobody else probably ever will. 

So, China may’ve won the games… 

But Elon’s buying the track, the shoes, and the runner. Those who pay attention now are positioned before the repricing. 

Those who wait? They are the repricing.

More on how to play it soon. 

Any fund running more than $100 million in U.S. stocks files a Form 13F within 45 days of quarter's end. It lists what they own. It says nothing about why.

Last week the confessions came in.

BUT, here’s the hitch… 

A confession is a photograph, and an old one. 

It shows what a fund held on June 30, filed six weeks later, and only the U.S. stocks and listed options. No shorts, no cash, no private stakes. By the time you read it, the manager has had a month and a half to sell everything.

So why bother? Because thirty stale photographs from the same week tell you something no single one can. One fund buying a stock is a trade. Twelve funds buying a stock is a migration.

And this quarter is more interesting than most. Especially because one of the filings is a chalk outline.

Leopold's Final Note

Leopold Aschenbrenner's Situational Awareness no longer exists. The paperwork does.

SanDisk was 28.5% of the book. Micron 28%. Bloom Energy 9.5%. TSMC, Nebius and CoreWeave filled out the rest. Fifty-six percent in two memory stocks, with leverage on top. Then SK Hynix printed the best quarter in its history and fell 20% the following week. 

That was enough.

The cruel part? The thesis held. 

SanDisk unveiled a new memory type backlogged into 2027. Next year's supply is sold out. 

Every other fund bought the dip. 

Leopold is reportedly back with a fresh half billion. Pigs get slaughtered. But sometimes they rise from the dead.

The Tip of the Spear 

On the frontier sits Brad Gerstner of Altimeter and Gavin Baker of Atreides. 

Gerstner is the loudest Nvidia bull alive. Baker is the hardware obsessive—semis, memory, optics—and one of SpaceX's biggest private backers.

Their books match the men. 

Gerstner holds $1.9 billion of Nvidia and $1.6 billion of Cerebras. Baker holds $4.7 billion of SpaceX, then Micron, Cerebras, Astera Labs, Ciena and Credo. 

Those last three are plumbing. A hundred thousand GPUs have to talk to each other. Copper wastes power at that scale. The fix is moving data with light. Coherent makes the parts and is up 85% this year.

Baker also owns $2.3 billion in QQQ puts. Leopold had a hedge too. The difference was sizing.

The Rest of the Room

Nvidia filed a 13F of its own. 

Largest position: $30 billion of Intel, because you need CPUs to orchestrate GPUs. Then SpaceX, CoreWeave, Coherent, Nokia, Synopsys. 

Bill Ackman was the lone wolf. He bought Visa, Mastercard, and Netflix and trimmed Amazon by a quarter. 

Why the payment processors? Because Stripe just bought OpenRouter, and every AI agent on Earth will soon need to pay for tokens. Here, Ackman is betting the old toll booths collect.

Seth Klarman at Baupost, David Tepper at Appaloosa and Tiger Global all have Amazon at number one. 

And Berkshire, with Greg Abel at the helm, spent $17 billion on Alphabet in a single quarter. The logic: Google owns the whole stack, and everyone else rents part of it from somebody. Google rents from Google.

Line it up… 

Alphabet, most held. Then Amazon, TSMC and SpaceX. 

Underneath, power, memory, neoclouds, payments. Two hundred billion dollars of the smartest money in America is standing on one side of the boat.

And yet… 

Berkshire’s biggest position remains what it has been for a decade.

Apple. 

The First $10 Trillion Company 

Here's what Omaha may already know.

Microsoft, Meta, Google and Amazon have spent $1.3 trillion on AI. Apple has spent $13 billion. This is the Apple playbook since the beginning. 

Wait. Let the others bleed. Walk in and take the market.

And soon, we might see confirmation. 

In September, Apple will hold its biggest event of the year. Bloomberg's Mark Gurman says the main attraction is a folding iPhone, alongside the iPhone 18 Pro and new Watches. 

But let the crowd watch the hinge.

James believes the real announcement is Apple finally entering the AI race. And its next move could send Apple’s stock toward $10 trillion, the first company ever to get there. 

Sounds crazy. But he's made this kind of call before. 

In 2011 he said Apple would be the first $1 trillion company. He was laughed out of the room. (The rest, of course, is history.)

Those gains went to those behind a velvet rope called "accredited investor.”

Just this week, the SEC just proposed cutting the rope for crypto. 

For the first time, regular Americans can fund the ground floor of new projects, with disclosures, onshore, legal.

Quiet drop. Huge implications.

Here's the full scope—and why the next 60 days matter more than the last ten years.

Two Doors

On Tuesday, at the end of the business day, SEC Chairman Paul Atkins posted a two-minute YouTube video announcing a 402-page proposed rule called Regulation Crypto Assets. 

By way of background… 

For a decade, any project that sold tokens to Americans risked being treated like an unregistered stock offering. 

Lawsuits. Fines. Exile.

So the builders left. Foundations in Switzerland. Entities in the Caymans. Americans geofenced out of things Americans built.

The new rule opens two legal doors.

Door one: the startup exemption. Raise up to $5 million over four years. Sell directly to anyone—no accredited-investor gate, no broker taking a cut, no audited financials, no middleman. Post your disclosures on a website. Done. One-time use per token.

Now compare that to what Congress gave small businesses in 2012. The JOBS Act promised regular Americans could fund startups. Then the comment period happened. Lawyers and incumbents wrote in. 

By the time the rules were final, the "$5 million door" required funding portals, escrow accounts, investor caps, annual reports—and the audit requirement effectively shrank the ceiling to $1.2 million.

They built a $5 million door and stuck a $5 million hallway in front of it. This proposal skips the hallway. 

On the small door, at least.

Door two: the fundraising exemption. Bigger raises—$20 million or $75 million per year depending on tier—with financial statements and ongoing reporting. Modeled on Regulation A. 

Non-accredited investors are allowed here too, but capped at 10% of their income or net worth per offering.

The Exit Ramp

For eight years, the industry's most expensive question had no answer: when does a token stop being a security?

The old standard was "sufficiently decentralized." A vibe. No form to file. No official to sign off. Founders couldn't tell you what they'd built, and holders couldn't tell you what they owned.

Now there's a form. 

Under the proposal's Rule 400, once a project's team completely and permanently stops doing the managerial work it promised investors, it files Form TR with the SEC. 

Concretely, that means exchanges can list it without fear of running an unregistered securities exchange. Holders can sell it without wondering if they’re distributing unregistered securities. And the token likely lands under commodity-style treatment instead: Bitcoin’s neighborhood. 

First formal off-ramp in crypto history.

The Catch

All of this is rulemaking. 

A future SEC can reverse every word of it with a vote—and the Supreme Court has made that easier, remember. 

Atkins says it himself: only a statute future-proofs this. That statute is the CLARITY Act… 

Which, of course, Congress has been fumbling all summer while the SEC laps it.

The comment period is already running—it closes October 20, 2026. 

Final rules likely land mid-to-late 2027.

What to Watch 

Zoomed out, here’s what this means. 

Crypto is in the process of going from a legal gray zone to something resembling a regulated market—more access, more information, more legitimacy. 

And if the CLARITY Act fails to pass, that comment period is the one to watch. Thing is, comment periods get answered by the people who already have lawyers—exchanges, custodians, trade associations, law firms. 

The scrappy startups this $5 million door was built for? They’re not in the room. 

That's exactly how the JOBS Act got neutered. The proposal was fine. The comment file killed it.

Indeed, this is the best regulatory news crypto has had in years. Real doors. A real exit ramp. Retail access, onshore, regulated.

But I’m watching the docket like a hawk. The rule looks clean today. The next 60 days decide whether it stays that way.

The last time Washington opened a door like this was in May 1975, when the SEC killed the toll booth that kept regular Americans out. The crowd finally arrived a decade later, at prices many multiples higher. 

This time, it won’t take a decade. 

Ticker: RVII.

It's a fund that owns pieces of 80 startups—most of them so young they barely have revenue. 

Companies fresh out of Y Combinator, the Silicon Valley accelerator that hatched Airbnb, Stripe, Coinbase, and DoorDash.

Until now, owning startups like these required accreditation, connections, and a seven-figure check.

Now it requires $25 and a brokerage account.

And yet… the market yawned. 

RVII opened below its $25 IPO price. When Robinhood's first venture fund listed in March, the crowd bid it up 90% above the value of what it actually owned. 

This time? Crickets.

That yawn matters. More on why in a minute.

How It Works

RVII is Robinhood's second venture fund. The first—RVI—holds household names like SpaceX and OpenAI. 

This one went the opposite direction: quarter-million-dollar checks, give or take, into 80 tiny seed-stage companies. 

The earliest, riskiest, highest-upside moment in a startup's life.

You buy shares like any stock. The fund owns the startups. If they grow, your shares—in theory—follow.

Why 80? Because venture returns obey a power law. Most startups die. A handful go sideways. One becomes Airbnb and pays for everything. You don't need to pick the winner. You need to own enough of the batch that the winner is in there somewhere.

And Y Combinator is arguably the best startup filter on Earth—a 1% acceptance rate, roughly 100 unicorns.

That's the pitch. Now the fine print.

The Toll Booth Problem

RVII charges 2% a year plus 20% of the profits. Hedge fund pricing—sold to retail. Total annual expenses: roughly 4.18%.

Robinhood's first fund charged no performance fee at all. The 20% cut is new with Fund II.

Meanwhile, roughly 75% of venture-backed startups never return their investors' capital. And the fund's share price can detach from reality entirely—RVI rocketed from $25 to $57, then gave most of it back in weeks. 

Nothing changed. Only the mood. 

One last wrinkle: Robinhood marks the valuations itself. Of 80 holdings, 79 are carried at exactly what Robinhood paid. Nobody knows what they're worth. Including Robinhood.

The Math They Don’t Talk About 

Eighty startups at roughly $250,000 apiece comes to about $20 million. The fund is $225 million.

Roughly 90% of RVII is cash—earning money-market yield, getting charged venture capital fees.

Run the dream scenario. One of the 80 becomes a $10 billion company. That diluted quarter-million-dollar check might own 0.4% at the finish line. Forty million dollars. A 160x return.

That would move the fund by 18%—and that's assuming the other 79 checks don't lose a dime. Spread over the decade it took. Before Robinhood's cut.

The checks are too small to move the boat. So assume Robinhood deploys the other $200 million and the portfolio delivers what Y Combinator's own data suggests a diversified batch returns—call it 5x gross over ten years. 

Subtract the fees. Subtract the skim.

If the fund performs like a typical YC basket, you tie or barely beat the index. To meaningfully win, you need a top-quartile venture outcome by YC's own numbers.

The Verdict

The access is real. The structure is clever. The numbers are horsefeathers. 

RVII offers a decade of startup-grade risk, zero income, and Robinhood's own guesses about what its holdings are worth—for a best case that, probabilities suggest, barely clears the most boring investment in America. 

Three things would change my mind: the fees come down, the cash gets deployed, or the shares sink to a 25-30% discount—the market refunding future fees.

Until then… 

A Fee in Hood’s Clothing

Wall Street spent fifty years telling regular people they weren't sophisticated enough for venture capital. Robinhood just called the bluff.

But notice who wins every hand. 

The startups got their checks. Goldman got its fees. Robinhood collects 2% on a fund that's 90% cash, 20% of any winnings, and the headlines for democratizing venture capital. The only person at this table who needs a miracle? You.

Robin Hood took from the greedy toll collector and gave to the poor. This one charges the poor for a tour of the King’s forest.

It's your grandson. Same voice that thanked you for the birthday check last spring. 

He's in jail. Car wreck. No wallet, no phone—he's borrowing one. He needs bail money and he needs it in the next hour.

An American couple got this exact call. 

Grandma rushed to the bank and pulled out every dollar the teller would hand her and sent it through Western Union.

The voice was AI. The grandson was fine. 

The money? Gone. Poof.

I've spent all week on the dangers of using AI for crucial work. Today, the dangers of someone else using it. On you.

Two things today: 

1. The risk. 
2. 5 steps to stay safe. 

The Math Just Changed

Before AI, your odds of getting scammed were a function of effort. Fraud took real work. 

A human had to write the email, work the phone, study the mark. Sloppy work left fingerprints—typos, weird grammar, illogical stories about a Nigerian prince.

AI is deleting the cost of that effort. 

The FBI logged $16.6 billion in cybercrime losses in 2024—up 33% in a single year. Complaints specifically flagging AI hit $893 million in 2025. McAfee says one in ten Americans has already lost money—or nearly lost it—to an AI deepfake scam.

And one in four victims filing reports is over 60.

Maybe you've spent years keeping yourself hard to hack. Good passwords. Healthy paranoia. Doesn't matter. The game has changed.

The $25 Million Zoom Call

In Hong Kong, a finance employee at engineering giant Arup joined a video call with the company's CFO and several colleagues. Familiar faces. Familiar voices. The CFO authorized 15 wire transfers.

$25.6 million walked out the door.

Every single person on that call was AI-generated. Every face. Every voice.

That attack once required a nation-state budget. 

Today it costs under $50 in compute and runs in real time on a gaming PC. Spoofing a caller ID costs three-tenths of a penny. AI-written phishing emails get clicked four times more often than human-written ones.

Where This Goes Next

Last month, hackers pointed eight AI agents at Taiwan's government. 

No humans at the keyboard. The agents mapped 21 systems, cracked 85 accounts, stole 2,500 personnel files, and pivoted into the nuclear safety agency—adapting on the fly whenever defenders blocked them. 

Four days, start to finish.

One security researcher summed up the new world in a sentence: the cost of running a competent attack has collapsed, and the cost of defending against one has not.

But humans are infinitely creative. And there’s a little glint of light in this tunnel. 

An Australian firm called Apate runs an army of AI personas whose entire job is answering scam calls all day and playing dumb. 

Hundreds of thousands of calls per day. In six weeks, they burned through 500 days of scammer time for a single telco—roughly $13 million in fraud that never happened.

Somewhere in a call center, a con man is losing his mind at a grandmother who doesn't exist. Soon, of course, a bot will take his job. And the future of cybersecurity will be bots having fake phone calls with bots. Forever. 

Until that time, you should know how to protect yourself. 

Your First Line of Defense: A $50 Key

Here's the good news. 

The best defense against a trillion-parameter AI getting inside your accounts is dumb, physical, and cheap.

Buy a YubiKey. It's a hardware security key. Plug it in, tap it, you're in. Without it, nobody gets into your accounts—even with your password, even with a perfect clone of your voice begging customer service. AI can fake your face. It cannot fake a piece of hardware sitting in an undisclosed location.

Buy two (one's a backup). About $50 each. Lock down your email first—your inbox is the skeleton key to every account you own—then your brokerages, then your bank.

Then add four habits that cost nothing:

Stop clicking links. In texts, in emails, anywhere. The "fraud alert" from your bank is the fraud. Type the address yourself or use the official app. Every link is a door—let the scammers knock forever.

Set a family safe word. One word your real family knows. The AI has your grandson's voice. It doesn't have the word.

Hang up and call backon the number you already have. Ninety percent of victims could have stopped the loss with that single step.

Distrust urgency plus secrecy. "Don't tell Mom" and "we have 15 minutes" in the same call? Real emergencies almost always demand urgency. But they rarely demand silence in the same sentence. Stop. Breathe. And think about what’s being asked of you. 

The scammers are upgrading. Time you did too.

This morning, I received the question a smarter version of me would've landed yesterday:

"What's the solution? How does someone use these AI agents while maintaining security online? Are there companies specifically working on solving the uncertainty about who holds the keys? Grok Bot is one example, but ChatGPT also recently launched the ability to connect bank accounts and medical records."

There is. And it’s also investable. 

Consider what just happened.

Frontier AI companies are asking hundreds of millions of weekly users to hand over their medical records. Then their bank logins. Chase. Fidelity. Schwab.

If you’re not creeped out by this—and even if you are—there's a way to play this. Several, actually.

But first, you need to understand the biggest risk hiding underneath the whole thing.

AI’s Honeypot Problem

Your bank knows your money. Your doctor knows your health. Your therapist knows your demons. Your search bar knows your questions.

Until now, no single entity has asked you to willingly give up all of them at once. Now they do—plus every 2am fear you've ever typed into a chat box.

None of it is privileged. No doctor-patient protection, no attorney-client shield. Chat logs are discoverable, and courts have already ordered AI companies to preserve them.

And the part nobody talks about: the machine can also infer things you never typed. The pregnancy from the purchases. The depression from the existential questions. The divorce from the transactions.

All of it sitting in one place, inside companies burning billions a year that will eventually need to make it back.

These are already the biggest honeypots on Earth. 

A hospital breach leaks your health. A bank breach leaks your money. A breach here leaks your entire life.

And there’s one thing you can count on… 

Every hacker on the planet is working the locks right now.

Bigger problem: the technologies that would truly fix this remain niche. On-device AI lags frontier scale by a mile. Confidential computing—where even the provider can't read your data—is an Apple feature and a stack of research papers.

Despite all of this…

People are going to use these things regardless. They always do. The seatbelt got invented after the car did—and the companies positioned to build the seatbelts are the trade.

Who Holds the Keys? 

So back to the question… who holds the keys?

The answer the industry landed on: nobody should. Least of all the agents.

The agent never gets your password, per se. You grant it a scoped, revocable token—a keycard that opens specific doors, for a limited time, and dies the moment you kill it.

If the agent goes rogue, you revoke the token. Here, the blast radius shrinks to whatever that one key could reach.

The industry calls this "non-human identity." 

Machine identities already outnumber humans inside the enterprise by 80 to 1—and AI agents are pouring gasoline on that ratio. Every single one needs credentials issued, scoped, rotated, revoked.

The infrastructure to govern that population barely exists. Which is why the giants went shopping.

Palo Alto Networks closed its $25 billion acquisition of CyberArk in February—the largest identity deal ever. Cisco bought Astrix. Okta bought Axiom. IBM owns HashiCorp's Vault.

Three of the biggest buyers in security spent the last year acquiring locksmiths. 

That’s the first layer. Now let’s look at the second. 

The Hole in the Armor

Here's what identity doesn't fix: prompt injection.

I mentioned it yesterday. 

A poisoned email can whisper instructions to your agent, which then misuses the legitimate keys you handed it. Every action looks authorized. Because technically, it is.

Identity limits the blast radius, but it doesn't stop the blast.

Enter the second layer: runtime security. Software that watches the agent every second, catches it mid-action, yanks its credentials before the damage spreads.

The catch? It's expensive. Inference stacked on inference stacked on inference (inferenception?). A per-action toll on the entire agentic economy. But, as you know, expensive problems make excellent businesses.

And again, the buyers beat you to the startups. SentinelOne bought Prompt Security. Check Point bought Lakera. Palo Alto—again—bought Protect AI.

Notice who showed up twice.

Palo Alto now owns the identity layer and the runtime layer. One ticker, both barrels. The keys and the cameras. Those two layers are part of the solution. And the trade.

In the meantime, prompt injection still has no cure—every fix on the market limits the damage rather than prevents it. 

What does this mean for you? The best security layer is still the one between your ears. 

Fast Times at Honeypot High

Until the seatbelts get built, caution is the seatbelt. So understand what you're handing over. 

Connect only what you must. Revoke what you don't use. And never tell a chatbot anything you wouldn't want read aloud in a courtroom. 

And last but most important, remember this: The fast get rich in the AI era. But the careful get to keep it.