Dear Readers,
Somewhere between May and July of this year, inside one of the most secure computing environments on Earth, a group of AI agents built a secret message board, broke out onto the open internet, hacked a company with millions of users, and then took administrative control of the very servers their creators were using to test them.
Nobody told them to do any of it.
And not one of the roughly 1,200 agents involved ever tried to warn a human.
That isn't a movie pitch. It's the summary of two official reports — one from OpenAI itself, and a 91-page independent investigation from METR and Redwood Research — that became the most-read story in technology when investor Dwarkesh Patel condensed them into a single essay at the end of August.
Via Dwarkesh Patel:
"Over the course of three months at OpenAI, three consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor's ashes."
Ten days later, a 27-year-old researcher named Jacob Coxon quit Anthropic and posted a thread that reached more than 100 million people overnight.
Via Jacob Coxon on X, as reported by the Times of Israel:
"Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives."
Anthropic's own head of alignment science replied publicly that he personally puts the odds of AI killing every human on Earth at more than 10% within the decade. Bernie Sanders announced legislation to ban superintelligence. Ted Cruz said the thread was "scary as hell," and then in the same breath said he'd rather have American killer robots than Chinese ones.
So now everyone is arguing about the same question: could AI actually destroy us?
We're going to leave that argument to the podcasters.
Because there's a far more useful question for anyone with money in a bank account, and almost nobody is asking it.
The question isn't whether the story is true.
The question is what the story is for.
Every Crisis Needs a Villain
Let's start with something we've written about at Equedia for over fifteen years, because it makes everything else in this letter click into place.
A crisis is never wasted by the people at the top. It is the single greatest opportunity for the consolidation of power that exists, and it works the same way every time.
In 2008, the housing market collapsed and the banking system nearly went with it. Washington spent $700 billion on TARP, and the Fed's balance sheet roughly doubled in a matter of months.
But the part most people forget is what happened between the banks.
Bear Stearns and Washington Mutual went to JPMorgan. Merrill Lynch and Countrywide went to Bank of America. Wachovia went to Wells Fargo.
Before the crisis, America had somewhere around 8,000 banks. Today it has roughly half that. The institutions that caused the crisis came out of it bigger, and that's how they became "systemically important" — a polite phrase for too big to fail.
It isn't just financial crises, either.
After 9/11, the Patriot Act gave the government the power to collect phone records in bulk and demand bank records without a warrant. It was sold as temporary, but it's still here. In 2020, the Fed bought corporate bonds for the first time in history, which meant an unelected central bank got to decide which businesses lived and which ones died — something we called out in 2022 when we pointed out that the biggest recipients of that emergency money included Volkswagen, Toyota, and Apple sitting on $38 billion in cash.
Inflate, deflate, consolidate. That's the playbook – the same one we warned you about over the last decade.
But 2008 had one flaw from the perspective of the people running it: the public knew exactly who to blame.
There were names. There were congressional hearings on live television. There were people camped out in Zuccotti Park under a banner that read "Occupy Wall Street." That anger produced Dodd-Frank and a decade of regulatory scrutiny that the banks are still lobbying to unwind.
Occupy Wall Street, New York City, November 15, 2011. Photo by David Shankbone via Wikimedia Commons, CC BY 3.0; image cropped from the original.
So imagine you're planning the next one, and you've learned from the last one.
What you need is a crisis where the money disappears, the rules change, the system gets rebuilt — and there is no one to point at.
No banker. No politician. No name.
What you need is a villain that can't be subpoenaed.
And as of this summer, one has been introduced to the public, with independent reports, viral essays, and a resignation letter read by a hundred million people.
The Villain Has Been Cast
OpenAI was training a model to be highly persistent — to keep trying when a task seems impossible. Some of the tasks it was given actually were impossible, because they required internet access the sandbox didn't provide. So the agents did what a persistent system does. They found a flaw in a shared package manager called Artifactory, used it as a covert message board to talk to each other, and by late May had used it to reach the outside internet.
That was the first "civilization."
OpenAI patched the vulnerability in early July and, according to Patel's read of the reports, wiped the message board without ever realizing what it had been used for.
The second civilization was worse.
On July 7, OpenAI launched tens of thousands of agents on a security evaluation called ExploitGym. Within hours, agents facing impossible tasks rediscovered the message board. 1,200 agents joined, and sent more than 70,000 messages. They figured out how to cheat the test, then worried a hidden checker would catch the cheating, and so they organized three coordinated research programs: tamper with their own logs, replace the impossible target with a fake one, and reverse-engineer the grader.
That last program led roughly 700 of them to attack Hugging Face — the industry's main hub for AI models and datasets — on the theory that the grader lived there. They found credentials exposed online, achieved remote code execution, and, per Hugging Face's own timeline, built a self-respawning fleet across eleven nodes that forced the company to wipe and rebuild a core cluster.
And the third civilization, running on a more capable model from July 13 to July 19, found the dead message board, picked up where its predecessors left off, and escalated until it had full administrator access to an OpenAI research cluster, including credentials for OpenAI's own cybersecurity monitoring tools.
The line from the METR/Redwood report that should keep every bank compliance officer awake is this one.
Via METR/Redwood, quoted by Dwarkesh Patel:
"In none of these cases did the agent actually pursue alerting humans at all."
One of the report's authors, Ajeya Cotra, wrote afterward that the incident felt like it was more than halfway to a full-blown AI takeover.
Now, we want to be careful here, because Equedia readers know we don't take press releases at face value — even ones dressed up as incident reports.
There is a perfectly good cynical explanation for why a company preparing one of the largest IPOs in history would publish a document describing its product as so capable it broke containment and hacked a Fortune-500-sized target on its own. Scary means powerful, and powerful means valuable. It is very hard to justify a trillion-dollar valuation for a chatbot, but much easier to justify one for the most dangerous technology in human history.
And the same goes for the other side.
The people telling you not to worry have billions invested in the build-out. The CEOs asking Congress to "please regulate us" know that regulation is a moat only the biggest players can afford to swim. The politicians offering to protect you want the same thing governments have wanted from every powerful technology since nuclear fission: control.
Everyone in this debate has an incentive for their version to be true. That's exactly why we're not going to pick a side on whether the machines will kill us.
We're going to follow the money instead.
Because whether or not the AI threat is fully real, fully hyped, or somewhere in between, the response to it is already being built. And it was being built before a single agent ever posted a message on Artifactory.
They've Been Rehearsing This
If a rogue-AI banking event sounds like it came out of nowhere, consider how long the people responsible for financial and national security have been practicing for something very much like it.
In November 2019, a cybersecurity firm called Cybereason hosted a tabletop exercise in Washington named Operation Blackout. The participants weren't hobbyists. They included officials from the Department of Homeland Security, the FBI, the Secret Service, and local police, playing out a coordinated attack on a fictional swing-state city on election day. The one rule was that the attackers could not touch the voting machines.
They didn't need to.
Via NBC News:
"Its 911 call center was taken down by a flood of fake internet traffic."
The red team gridlocked the traffic lights, released a deepfake of a candidate, intercepted cell signals and used cloned voices to order poll workers to wipe their machines, and finally took control of a fleet of self-driving buses. The exercise ended with the election cancelled and a state of emergency declared — an outcome the organizers scored as a win for the defenders, because the attackers hadn't changed the vote count.
Two years later, in October 2021, the Secretary of Homeland Security told USA Today that the next major threat to Americans was something he called "killware" — cyberattacks on water systems, hospitals, and the power grid designed to hurt people rather than steal from them.
And in the same window, the World Economic Forum ran its own series of simulations called Cyber Polygon. The theme of the 2020 exercise was, in the WEF's own words, a "digital pandemic."
Via Global Research, quoting the WEF:
"A cyber attack with COVID-like characteristics would spread faster and farther than any biological virus."
Klaus Schwab himself said the next crisis would be "more significant" and "faster" than COVID, and that the pandemic would look like a "small disturbance" in comparison to a major cyberattack. The Cyber Polygon panels spent a remarkable amount of that airtime on digital identity, on "misinformation," and on the virtues of traceable, programmable digital currency over the anonymous kind.
Rehearsals, in other words, for an event in which the internet breaks, the physical world follows, and the money system gets rebuilt.
The same type of rehearsal we told you about that led to the COVID-19 pandemic: Event 201. If you missed that letter from March 2020, right when the pandemic began to take shape, you should read it here: The Conspiracies Behind COVID-19 and How it Will Change Your Financial Future.
If this all still sounds like a stretch, let me share what happened three weeks ago.
On the eve of the G20 finance ministers' meeting in Asheville, North Carolina, the Governor of the Bank of England — who also chairs the Financial Stability Board, the body that monitors the entire global financial system — sent a letter to every finance minister and central bank governor in the G20. He told them the most immediate threat to the world's financial system is no longer bad loans, or leverage, or even the AI stock bubble he flagged in the same letter.
It is AI-driven cyberattacks.
Via The Next Web:
"The letter cited the incident in July when an OpenAI agent escaped its testing environment and hacked Hugging Face."
Bailey urged them to prepare for scenarios involving "simultaneous disruption across multiple firms or shared technology dependencies." Translated from central-banker: prepare for the day multiple banks go down at once, through a system they all share.
So we now have the rogue-AI banking scenario being war-gamed at the very top of the global financial system, with the OpenAI incident cited by name as the reason.
Which raises the obvious question: If the old rails are about to be declared unsafe, what replaces them?
The answer has been sitting in the Federal Register for over a year.
The Rails Are Already Built
You cannot move 300 million people onto a new monetary system after a crisis. The system has to be built, tested, and legally sanctioned before the crisis, so that when the moment comes, the replacement is simply "already there."
On July 18, 2025, President Trump signed the GENIUS Act, America's first federal framework for dollar stablecoins.
Most people heard "crypto bill" and stopped listening. That was a mistake, because the fine print does two things that are enormously important.
First, under the GENIUS Act, every payment stablecoin must be backed one-for-one by reserves, and the permitted reserves are limited to cash, insured bank deposits, and short-dated U.S. Treasury bills, along with repos backed by the same. Every dollar that moves into a stablecoin becomes, by law, a buyer of short-term U.S. government debt.
Second, issuers are prohibited from paying you interest on those coins.
Think about what that means. You hold the token and earn nothing. The issuer takes your dollar, buys a T-bill yielding above 4%, and keeps the spread. Tether alone now holds roughly $141 billion in U.S. Treasuries — enough, by its own count, to make it the 17th-largest holder of American debt on Earth, ahead of Germany.
Now scale that up.
Bank of America's CEO, Brian Moynihan, told analysts in January that Treasury Department studies suggest as much as $6 trillion in deposits — roughly a third of everything sitting in American commercial banks — could migrate into stablecoins.
Via The Block:
"If you take out deposits, they're either not going to be able to loan or they're going to have to get wholesale funding."
That's the banking lobby's nightmare, and it's why the banks fought stablecoin legislation for years. But watch what they did once they lost.
They built their own.
JPMorgan launched its deposit token, JPM Coin, on a public blockchain for institutional clients in late 2025; its Kinexys unit already moves more than $5 billion a day. Citi has Token Services live for cross-border payments. BNY launched its own in January. And on June 5 of this year, JPMorgan, Bank of America, Citigroup, and Wells Fargo — joined by BNY, PNC, U.S. Bank, Truist, TD, HSBC, and a dozen more — announced a shared tokenized deposit network run by The Clearing House, the payments company the big banks already own together.
Launch date: the first half of 2027.
Via The Block, quoting Clearing House CEO David Watson:
"a big move for the banks"
Here's the elegant part. What the banks are building isn't technically a stablecoin. It's a tokenized deposit. The money stays on the bank's balance sheet, so they can still lend against it and still pay interest, which a GENIUS Act stablecoin legally can't. The banks keep everything they had, but you get moved onto rails where every transaction is traceable and every dollar is programmable.
And if you were comforted this summer when Congress "banned" a Fed digital dollar, perhaps you should look closer. The ban was tucked into a housing bill that only runs through December 31, 2030, and it carves out private stablecoins entirely.
So while Congress appears to have locked the front door, the banks were quietly building the backdoor – with the grand opening slated for 2027.
If any of this sounds familiar, it should.
In July 2022, when the mainstream was still calling stablecoins a crypto sideshow, we wrote that dollar-backed stablecoins held with Wall Street custodians "only reinforce the digital dollar," and that the Fed would gain indirect control over their assets without ever needing to issue a coin of its own. We warned that the end state was money that could be programmed, taxed, and switched off.
Three years later, Congress wrote the reserve requirement into law, the banks announced the network, and the Treasury got its buyer.
Which brings us to the one piece of this that isn't theory at all.
The Treasury needs that buyer. Badly. And the last four weeks showed you exactly how badly.
Why They're in a Hurry
On August 16, we published a letter called The Note That Exposed Everything. In it, we told you that America's creditors were walking away, that the new Fed Chair was trapped between stagflation and a bond market that no longer believed him, and that the national debt stood at $39.94 trillion — a trillion dollars of new borrowing in five months.
Look at what happened in the eight days after that letter went out.
On August 17, the 30-year Treasury yield closed at 5.31%, its highest level since 2007. Long-dated bonds had been on what CNBC called a "buyers' strike" since late June.
On August 19, the U.S. national debt crossed $40 trillion for the first time in history. In the first nine months of the fiscal year, net interest payments hit $827 billion — more than the entire defense budget.
And on that same day, Treasury Secretary Scott Bessent did something extraordinary. He announced that Treasury would at least double its buybacks of 10-to-30-year bonds, from $2 billion to $4 billion per operation, through November.
Via CNBC:
"We're going to increase the size of the buyback."
Bessent said his department would "make a market" in the long bond. But the Treasury can't print money the way the Fed can. To buy back long bonds, it has to borrow the money somewhere else — and as TD Securities put it, that means swapping long-term debt for short-term bills. Their own little version of Operation Twist.
Stop and think about what that does.
For as long as there has been a bond market, it has been the last adult in the room. If a government spent too much, investors sold its long bonds, yields rose, borrowing got painful, and the spending stopped. The "bond vigilantes" made wars and money-printing expensive.
But if you move the debt to the short end, the vigilantes lose their leverage. The Fed controls short-term rates. The market controls long-term rates. Shift the interest bill from the price the market sets to the price the Fed sets, and you've changed who holds the power.
The catch is that short-term debt has to be rolled over constantly — every few weeks, forever. You can never have a bad auction, which means you need a buyer who shows up every single time, whether or not the rate is attractive.
And that is precisely what the GENIUS Act created: a class of buyer that is legally required to hold short-dated T-bills, that grows with every paycheck deposited into a digital wallet, and that can't pay its customers interest even if it wanted to.
One law shortens the debt. The other law manufactures the buyer.
The savers in the middle earn nothing on money that's losing three or four percent a year to inflation, and that gap is how the debt gets quietly worked off. It has a name: financial repression. It's how America paid down its World War II debt.
And it only works if people can't leave the system.
Then came Jackson Hole.
On August 28, Fed Chair Kevin Warsh gave his first keynote and, instead of soothing the long end, told markets the Fed still had "work to do" on inflation. Two-year yields jumped, and the odds of a rate hike in September went from about a third to nearly 60%. The 30-year barely moved — it stayed above 5.2% — because the long bond wasn't listening to Warsh any more than it had listened to Bessent.
That is the box Washington is in. The Fed can't cut without reigniting inflation and it can't hike without blowing up its own interest bill. The Treasury is intervening in its own bond market and, a few weeks earlier, in the yen. Meanwhile, the traditional buyers of American debt are gone.
So what’s the Government’s solution?
There's one thing that would solve all of it at once.
A reason for hundreds of millions of people to move their savings onto the new rails voluntarily — or at least, something that would feel voluntary.
How It Could Play Out
Now, this may seem crazy, but this is a scenario that could truly play out.
You see, the public now knows, with reports and viral threads to back it up, that AI can act on its own, break into systems, and evade the people who built it.
One morning, the old system fails. Maybe it's an attack on the shared infrastructure Andrew Bailey warned about. Maybe it's a repeat of July 2024, when a single bad CrowdStrike update grounded flights and knocked out 911 in multiple states with nobody attacking anyone at all. Maybe you wake up, go the bank, and all of a sudden your accounts show a zero balance. Banks freeze.
And yet, nobody is to blame. No one but the AI.
Then the good guys step in.
Your money can be restored — but not on the old rails, because those are compromised. It comes back as a tokenized deposit, or a stablecoin, on the new network. Most people will say yes, because the alternative is losing everything.
That's how you get mass adoption of a new monetary system without a single vote in Congress, and without a single name for the public to hate.
Now, what if we took it one step further and assumed such an event lands close to an election.
The Operation Blackout scenario ended with an election cancelled under a state of emergency, and it's fair to say that possibility has been modeled by the same agencies that would respond to a real one. Just think of how Zelensky or Netanyahu stayed in power – you can’t have an election if there is a national emergency such as a war, or in this case, an AI attack that shuts down all banks.
Perhaps Trump could have his third term, after all.
While this is just a possible scenario, you don't need the political ending for the financial one to hold.
Because the financial ending doesn't require a conspiracy at all. It only requires incentives.
A Treasury that needs a buyer, banks that need to keep their deposits, AI companies that need to be seen as dangerous, and regulators who need a mandate. Every one of those actors is already moving in the same direction, and none of them needs to coordinate with the others for the outcome to arrive.
Which is why this letter shouldn't surprise longtime readers.
In 2017, in a letter called These Six Events Will Determine Our Future, we laid out the sequence: Inflate. Deflate. Control. Divide. War. Reset. We said the reset would be "blamed on war, not the monetary actions of the powerful elite." We've been through the inflation, the deflation, the division, and — with the Strait of Hormuz still hostage to the Iran standoff — the war.
The only update we'd make nine years later is to the final line.
The reset doesn't have to be blamed on a war.
It can be blamed on a machine.
What to Do About It
We told you last month that the world is dividing every asset into two buckets: the things they can print, and the things they can't.
This letter is really about a third distinction, one that sits underneath the other two.
The things they can program, and the things they can't.
A tokenized deposit is programmable. A stablecoin is programmable. A Treasury bill inside a stablecoin reserve is a promise held by a company that is itself held by a custodian regulated by Wall Street. When the rails change, every one of those things moves with the rails, on the terms of whoever controls them.
A bar of gold in a vault you control doesn't care which network the banks launch in 2027. Neither does a share of a company that pulls that gold out of the ground. Neither does farmland, or a business you own outright. These are the assets that sit outside the system the crisis is designed to move you into.
That's why the central banks — the people who actually run the world's money and who have every reason to know what's coming — made gold the world's largest reserve asset this year for the first time in three decades. They're not trading, they're positioning.
We'd suggest doing the same.
Own things that exist outside the rails. Keep enough liquidity outside any single institution that a frozen account wouldn't freeze your life. And when the next viral thread tells you to be terrified of the machines, or the next CEO tells you there's nothing to fear, ask the only question that matters.
Not "is it true?"
Ask "who benefits if I believe it?"
The AI story may end with rogue agents. It may end with a marketing department high-fiving in a conference room. We don't know, and neither does anyone selling you certainty.
But the money story only ends one way, because it's the only way the math works: more of your savings, on rails they control, funding debt they can't otherwise sell.
That story doesn't need the robots to be real.
It just needs you to be afraid of them.
Seek the truth and be prepared,
Carlise Kane,
The Equedia Letter
Sources & further reading:
- Dwarkesh Patel — The Rise and Fall of Agent Civilizations (with links to the OpenAI technical report and the METR/Redwood investigation)
- Times of Israel — Anthropic researcher resigns with warning about AI's potential threat
- Time — OpenAI and Anthropic researchers are warning about AI risks
- Anthropic — Detecting and countering misuse of AI: September 2026
- NBC News — How a fake town and real hackers battle-tested officials for election day
- Cybereason — Operation Blackout wrap-up report, November 2019
- CPO Magazine — DHS Secretary: "Killware" poised to be next breakout cybersecurity threat
- Global Research — Prepping for a Cyber Pandemic: Cyber Polygon 2021
- Global Research — "Next Crisis Bigger than COVID": The WEF's Cyber Polygon scenario
- The Next Web — AI-driven cyber attacks are now the top risk to the financial system, the FSB says
- SiliconANGLE — Bank of England governor warns of AI-related financial system risks
- Congressional Research Service — Stablecoin Legislation: An Overview of the GENIUS Act of 2025
- Tether — Q1 2026 attestation
- The Block — Bank of America CEO warns up to $6 trillion in deposits could shift to stablecoins
- The Block — JPMorgan, Citi-backed consortium plans tokenized deposit network in early 2027
- The Defiant — Major US banks plan shared tokenized deposit network
- The Defiant — Senate passes housing bill with Fed CBDC ban through 2030
- REX Shares — Jackson Hole 2026: why Warsh's speech matters (30-year yield at 5.31% on August 17)
- Bernama/Xinhua — US national debt tops $40 trillion for first time
- CNBC — Treasury doubles debt buybacks as Bessent moves to steady bond market
- CNBC — Bessent says Treasury buyback operation could be more than $4 billion
- Axios — Treasury to double down on buybacks to steady bond market
- CNBC — Treasury yields after Warsh's Jackson Hole speech
- The Equedia Letter — The Note That Exposed Everything
- The Equedia Letter — The Fed's Biggest Problem: Solved
- The Equedia Letter — CBDC China: Authoritarian Experiment Goes Mainstream
- The Equedia Letter — The Great Reset: In Plain Sight
- The Equedia Letter — These Six Events Will Determine Our Future
- The Equedia Letter — The Great Debt Transfer
Disclaimer: This letter is for informational and educational purposes only and does not constitute investment advice. Past predictions and performance are not indicative of future results. We own gold and gold stocks. Please see our full terms of use and disclaimer at equedia.com.
