
I first published this on X on September 5th, 2026 – not to be prophetic, but this is more real than the date I published! In a Christian spirit, following the second of the greatest commandments from our Lord, loving my neighbors as myself, I do not intend to scare you, but to alert you that we need to ACT NOW to get a handle on AI/SI. We need established guardrails to protect our country and perhaps even the world from those who would use AI/SI against us in insidious ways!
I believe that a bank run no longer depends on a bad bank; instead, it just needs an elaborate AI fake and maybe a few hours’ notice at most. This scenario no longer seems hypothetical. If we keep the current gap between machine speed and human verification, someone — be it a criminal syndicate, a malicious service, or an organized group with a manifesto — will exploit it. They won’t have to break into a vault; they only need to appear as legitimate protectors of the vault.
Let me tell you what the next fourteen hours can look like. I am not describing a movie. I am describing a sequence for which we already have the pieces.
A document emerges after midnight, featuring a Board of Governors as if tailored to fit. It bears letterhead, type, watermark, and a hologram that your staff would initially hesitate to label fake. The headline states: systemic insolvency in major regional banks; an imminent liquidity crisis.
It is not a Fed release. Nobody at the FDIC cleared it. That will be true at hour one, and it will still be true at hour fourteen. It will not matter in time.
Within minutes, the file is everywhere, not because a newsroom authenticated it, but because other automated systems treat virality as a task. Then the confirmations arrive. Video of economists you would put on this dais. Audio of chief executives urging calm in language that quietly agrees with the panic. A voice that sounds like a compliance officer telling a treasurer to prepare for ring-fencing.
Those people did not record those clips. Some of them are asleep. Some of them are in other cities. But the public cannot tell. The night shift at a regional bank cannot tell when the phone rings and the voice on the other end has an authoritative cadence.
By dawn, lines form at banks and ATMs. Not because the banks are insolvent. Because enough people believe the next person will.
In 2023, Silicon Valley Bank lost $42 billion in just one day. This event involved a single franchise, primarily insured corporate cash, coordinated by people on message boards who still had to type. Some of you here have already been told that an agentic system could reduce such events from hours to seconds. I am now explaining the next level: not just one bank, but the *appearance* of the system itself.
Two trillion dollars in fourteen hours is not a fantasy number pulled from the air. It is what it looks like when fear is no longer local. Total deposits in this country amount to nearly twenty trillion. A fraction of that, moving at once, through apps that do not sleep, will freeze correspondent rails, empty ATM caches, gate money-market funds, and turn solvent institutions illiquid because liquidity is confidence with a clock on it.
The elderly couple who survived the 2008 housing crash won’t bother reading a forensic blog post at 6:14 a.m. They will simply get in their car to get their money. The payroll office won’t wait for your press conference. The hospital’s operating account won’t wait for a hash check. By the time the Fed, FDIC, and every listed bank release the same three statements *not our document, no insolvency, deposits are insured* the story is already unfolding. A denial becomes just another entry in the same feed as the forged documents.
We built deposit insurance to stop runs that started with a rumor in a lobby. We did not build it for a rumor that can be printed in the exact costume of the central bank, spoken in the exact voice of the CEO, and pushed by systems that do not eat, do not blink, and do not care.
Cybercriminals would love nothing more than the open doors they have.
They do not have to compromise core banking systems. They do not have to move a dollar out of a reserve account. They do not have to win a gunfight with a vault. They have to win a contest of appearances long enough for ordinary people to do their work for them. The payout isn’t just the chaos. The payout is the window — frozen rails, desperate wires, people who will believe the next voice that says, *I can get your money out.* Fraud rides in on panic as it has in every crisis we have ever had. Only now can the panic be manufactured on a schedule – using AI.
If criminals will do it for money, a state actor will do it for leverage. A blackout of trust is a weapon that needs no missile. You cannot sanction a PDF/computer file that never existed. You cannot freeze the assets of a voice clone and deepfakes. You can only discover, after the fact, that the public square can no longer tell a seal from a picture of a seal.
I have sat in rooms where a chief executive said, almost lightly, *What could go wrong? We are in control of AI.* I have watched another person, farther from the slide, answer: *Are you sure about that?*
I am here to tell you I am not sure. And if you are sure, you are not looking at the same clocks I am.
Controlling a model in a lab is not the same as managing what happens when related models can create official status more cheaply than authorities can prevent it. Guardrails mentioned only in terms of service are not true safeguards. A takedown issued after the issue has already emerged is not a policy; it’s a funeral notice for the morning.
I’m not asking you to stop new AI technology, and I can assure you our enemies aren’t either! I am asking you to stop acting as if “we are in control” is a proven fact. It’s simply a desire.
Treat the threat as real NOW! Forgery of official financial communications is a systemic-risk event, not a content-moderation dispute. When something wears the Fed, the FDIC, the OCC, or a chartered bank, the clock cannot be the same clock you use for a hoax about a celebrity. Anyone notice how real the hoaxes have become!? Even the Nigerian hoaxes and other scams are becoming harder to identify.
Require financial institutions and the agencies themselves to maintain out-of-band authentication that the public can actually use, and not a computer file PDF on a website competing with a prettier PDF and possibly AI-generated. A seal that a phone can check. A number that is not a screenshot. A human voice on a channel that cannot be cloned in the same hour, even if that means sounding less modern.
Bring together the platforms, model providers, and bank supervisors. The process no longer begins on a street; it starts in a feed. If an agentic AI can create a governor and a CEO before the open, then discussing safety and soundness without acknowledging that system is like talking about the last war.
Build the liquidity and communications drills for a *false* systemic story, not only a true one. Your stress tests still assume the shock is real. The shock that should keep you awake is the one that is fake and believed.
Do not wait for attribution. Attribution is how we comfort ourselves after the damage. The operators of a swarm will be a rented cluster and a story. By the time you know who they were, the ATMs are already dark and likely our banks.
I want to be careful, because fear without a point is entertainment.
No one can promise that a determined actor will fail. What you can promise is that we will not make it easy, cheap, or deniable. Right now, it is too easy, too cheap, and almost perfectly deniable. That is an invitation. Criminals read invitations.
Failure to act now, the next headline may sound like this:
*Ladies and gentlemen, what you are about to see has already triggered withdrawals exceeding two trillion dollars in the last fourteen hours.*
Your phones will not stop. The people in those lines will not care that you had a slide about AI alignment. They will care that the paper looked official and their life was in the account – and it is GONE!
I come here to sound an alarm. I came here because the cheap version of courage is to wait for the picture. I have seen the picture! It is a movie creepier than Bruce Willis’s Live Free, Die Hard with the Firesale! However, it may be a perfect fake: a night with no AI guardrails, and a country that still thinks it controls AI!
I believe we still have time to put a referee on the field and the resources to enforce the rules!
While what is in the article is fiction now (created well before the recent cyberattacks), the threat is very real, and we need to be proactive NOW. I have already offered up part of that “referee” here:

Leave a Reply