In One Paragraph

The Won't Get Fooled Again Act is a plan for the next financial crisis, written before it arrives. When a bank or a major AI company fails, the government does not bail it out or sell it to a bigger rival. It converts it into an institution with no shareholders: a bank owned by its depositors or by the public, a public research institute, or a public computing utility open to everyone. Shareholders, executives, and lenders take the losses. Workers, small suppliers, customers, and every depositor are protected. The industries that create the risk pay the costs, and the largest companies are barred from buying up the wreckage. If Congress waits until the crash, there is an emergency version: harder and more expensive, but still not a bailout.

What You Need to Know

The problem

• The American economy is leaning on an AI investment boom whose spending runs far ahead of its revenue, much of it financed with debt that reaches deep into banks, private credit funds, pension funds, and insurers.
• When bubbles burst, Washington has two habits: bail out the big players, and sell the failures to even bigger players. 2008 did both. So did 2023, when First Republic was handed to JPMorgan Chase, already the country's largest bank.
• Every rescue leaves the system more concentrated and more fragile, and spends borrowing capacity the country needs for real investment.

What the Act does

• Every failed bank that is still a functioning bank is converted over a weekend into a Public Benefit Bank, whatever its size: a bank with no shareholders, barred from speculation, and required to lend to the productive economy. Big banks get no better treatment than small ones.
• Local and regional banks become the property of their depositors, who elect the board, unless a state or city takes ownership. Only the largest, national banks with customers in every state, become federal public banks, and they also finance big national undertakings: the new industries and ambitious ventures that Wall Street passed over for speculation. The models are Germany's public savings banks, the Bank of North Dakota, and America's own mutual savings banks and credit unions.
• A failed bank is left unconverted only if the FDIC finds, in writing and in public, that it has no real base of customer deposits, no real lending business, or failed mainly because of fraud. Its accounts and sound loans then go to a nearby converted bank, credit union, or community bank. No failed bank's business may be sold to one of the giants, and the loophole that let JPMorgan Chase absorb First Republic is repealed.
• Aid means conversion. Any bank, and any systemically important technology company, that takes emergency federal support is treated as having failed and gets these terms. No federal agency, including the Federal Reserve, may offer a rescue on easier terms.
• A new agency (or an independent division of the FDIC), the FDIC-Tech, does for systemically important technology companies what the FDIC does for banks: places them in receivership when they fail, keeps their services running, and resolves them.
• From the moment a receiver is appointed, and for 90 days after, lenders may not seize or sell a failed firm's hardware, suppliers may not cancel its contracts, and counterparties may not call in its debts. Without that freeze, lenders would race each other to grab the hardware and finish the firm off.
• Failed AI labs become public research institutes. Failed data centers and cloud providers become a National Research Cloud, a public computing utility that sells capacity at regulated prices and gives affordable access to researchers, startups, nonprofits, government at every level, and the public at large.
• There is no ceiling on how much failed capacity the public may acquire, but there is a rule against hoarding: capacity left idle for more than 30 days must be offered for sale.
• Whether or not a crash ever comes, a fee on the largest AI and cloud providers funds Community Innovation Hubs in every congressional district: free community workspaces where people can start businesses, learn, and work together. Existing hubs can apply for the money as well as new ones. The models are the hundreds of incubators and coworking centers already operating in American cities, and public access television, which the cable industry paid for.

Who pays and who is protected

• Shareholders are wiped out completely. Executives responsible for a failure return five years of pay.
• Lenders are paid what their collateral is worth on the day of failure and no more. The government does not take on a failed company's debts.
• Once a lender has received the value of its collateral, workers, small suppliers, customers, and local governments are paid before banks and investment funds get anything more, by a priority list written into the law in advance.
• All bank deposits are insured, at every bank, with banks paying risk-based premiums for the coverage. A limit on insurance does not discipline banks; it sends money fleeing to the giants in every panic.
• Pensioners whose funds were invested in the failed loans are protected directly through an industry-funded Pension Protection Facility. The loans themselves take the loss.
• The public pays only for tangible assets and proven technology at current prices. It pays nothing for hype, goodwill, or growth projections.

Keeping the giants from getting bigger

• A national registry records who owns every large cluster of advanced AI chips, modeled on the aircraft registry. Owners need a simple license that must be issued to anyone who meets published security standards.
• No company that controls more than 5% of the nation's AI computing capacity may buy chips or data centers out of a bankruptcy or foreclosure. Nor may it lease chips bought that way, or buy a firm that has warned regulators it is close to failing. It can still buy as many new chips as it likes. The public gets first pick of the distressed assets. Everything else goes to smaller buyers.
• Chips from registered clusters are tracked through every resale, so any chips that disappear in a fire sale can be traced to whoever last held them.

Paying for it

• The technology side is funded by a tax on high-end chips and large training runs and by risk-based premiums from the largest technology companies. The banking side is funded by a surcharge on the largest banks.
• Because the funds will be nearly empty when they are first needed, they can borrow from the Treasury: up to $100 billion, repaid from future levies within 10 years, or up to $250 billion, repaid within 15 years, if the Treasury Secretary certifies that pending resolutions need it. The fund that capitalizes converted banks can borrow up to $75 billion on the same terms. This is how the FDIC's own fund is backstopped today.
• In a serious crash the debt of failed AI-related companies might run to $150 billion to $285 billion. Paying only for what the assets are worth, the public's outlay would be closer to $50 billion to $100 billion if used hardware loses most of its value in the crash, and perhaps double that if it holds up. Either way the public owns working assets worth what it paid. These are rough estimates.
• On the banking side the FDIC's records give the scale. The failures of 2023 would have needed about $43 billion to $53 billion of capital, and those of 2008 through 2013 about $55 billion to $69 billion over six years. That capital stays inside working banks. If a giant fails, its own loss-absorbing debt comes first.

If we wait until the crash

• The Act is designed to be passed in advance. Then the industry has paid premiums for years, the chip registry exists, the important firms have filed plans for their own conversion, and lenders knew the rules when they lent.
• If Congress waits, none of that exists, so there is a second, complete version of the plan for a crash already under way. It leans on the same rule this plan has, that any firm that needs public money to survive has failed, and applies it to firms that never prepared for it.
• In that version the whole cost is borrowed from the Treasury up front and repaid by the industry afterward; the FDIC acts as receiver from the first day; the dominant firms barred from buying distressed assets are identified in the bill itself; the rules take effect the day the bill is introduced, and sales already made to the giants can be unwound; failed companies already in bankruptcy court are moved into the new process; and every deposit is guaranteed at once by Act of Congress.
• Waiting costs more. Capacity will have gone dark, staff will have scattered, assets will already have been sold, the legal challenges are stronger, and the public advances all of the money. It is still far better than the alternative, which is a rescue written by the people being rescued.
• The emergency version is a full bill, not an outline. In 2008 the first draft of the bank rescue was three pages long and read as a blank check. In a crisis, whoever arrives with a finished text wins.

What it rests on

• A technology company can be placed in receivership only when it has actually failed, with a court able to review the decision within 24 hours. Three events count automatically, with no vote: filing for bankruptcy, taking emergency federal support, and leaving a payment of $100 million or more unpaid when its grace period runs out (30 days at most). Acting earlier than that is a judgment, and no single official can make it: it takes a written recommendation from two-thirds of the FDIC-Tech's board and two-thirds of the Federal Reserve Board, and a determination by the Treasury Secretary.
• Secured lenders receive the value of their collateral and every creditor receives at least what a liquidation would have paid, as the Constitution requires. Courts have the final word on value.
• The converted technology companies and the largest converted banks are federal public corporations with no shares, like the Tennessee Valley Authority, so there is nothing for a later administration to sell. They can be privatized only by an Act of Congress. Smaller converted banks belong to their depositors, or to a state or city.
• The plan does not promise that the federal boards are beyond a President's reach. Since the Supreme Court's 2026 ruling in Trump v. Slaughter, Congress may not be able to guarantee that. What protects these institutions is what the law fixes: their mission, the bar on privatizing them, and ordinary supervision. The plan is written for an administration that wants it to work.
• The precedents are American and familiar: the FDIC's weekend bank resolutions, the conversion of the bankrupt northeastern railroads into Conrail, the Tennessee Valley Authority, and the Bank of North Dakota.

The One-Page Summary

The United States is heading into its next financial crisis with the same playbook that failed in the last two. A large share of the economy's growth now depends on an AI investment boom whose spending far outruns its revenue. Much of that spending is borrowed, through private credit funds, off-balance-sheet shell companies, and loans secured by computer chips that lose value quickly. The debt reaches into banks, pension funds, and insurers. If the boom ends badly, the damage will not stay inside the technology industry.

When that happens, Washington's instinct will be what it was in 2008 and 2023: rescue the largest players with public money, and sell the failures to even larger ones. Each round leaves fewer, bigger banks and companies, more dependence on them, and less borrowing capacity for the things the country actually needs to build.

The Won't Get Fooled Again Act replaces that playbook with a simple rule: when a bank or a critical technology company fails, it is converted, not rescued and not sold to a bigger rival. A failed bank of any size, so long as it is still a functioning bank, becomes a Public Benefit Bank. It has no shareholders, and it is limited to the ordinary business of taking deposits and making loans to the productive economy: local businesses, housing, and infrastructure. Local and regional banks pass to their depositors, who elect the board, or to a state or city. The largest become federal public banks and also finance the big national undertakings and new industries that speculation crowded out. A failed AI lab becomes a public research institute. Failed data centers and cloud providers become a National Research Cloud, a public utility that keeps services running, sells computing power at regulated prices, and opens affordable access to universities, startups, nonprofits, government at every level, and the public at large. Separately, and starting the day the Act takes effect, a fee on the industry funds Community Innovation Hubs in every congressional district: free places to work, learn, and start a business, far from Silicon Valley.

The people who financed the bubble bear its cost. Shareholders lose everything. Executives return their pay. Lenders are paid the value of their collateral on the day of failure, not the amount they lent against it, and the government never takes on a failed company's debts. Workers, small suppliers, customers, and local governments are paid ahead of the banks and funds. Every bank deposit is insured, with banks paying for the coverage, because a limit on insurance only sends money running to the largest banks whenever there is a scare. Pensioners exposed to the failed loans are protected directly, not by rescuing the lenders who hold their money.

A crash is not allowed to make the giants bigger. A light national registry and license for large clusters of AI chips lets the government see who owns the country's computing power. Any company that already controls more than 5% of it is barred from buying distressed chips and data centers. The public has the first option, and the rest goes to smaller competitors. The rules are bright lines with deadlines, administered by the same agency that handles failures, so they do not become a new bureaucracy that says no to everything.

The industries that create the risk pay for the system. A tax on high-end chips and large training runs, premiums from the largest technology companies, and a surcharge on the largest banks fund the conversions. A Treasury credit line, repaid from those same levies, covers the early years. The public pays only what tangible assets are worth on the day of failure and nothing for hype. What comes out the other side is working banks and working data centers, owned by the public or by the banks' own depositors, and worth what was paid.

The time to pass it is before the crash, but there is a plan for afterward too. Passed in advance, the Act has years of premiums behind it, a registry of who owns what, and lenders who knew the rules. If Congress waits, a second version of the plan is ready for a crash already under way. It applies the same rule to firms that never expected it: any firm that needs public money to survive has failed, and gets these terms, whoever it asks, including the Federal Reserve. Acting late means borrowing everything up front, working without preparation, unwinding sales that have already happened, and fighting harder lawsuits. It is harder, more painful, and more expensive. It is still better than a third bailout.

The Act is insurance. If no crash comes, the conversions never happen, and what remains is a safer deposit system, a clearer picture of who controls the nation's computing power, community hubs across the country, and an industry paying for its own risk. If a crash does come, the country will be ready to turn the wreckage into public infrastructure, as it did when it built the Tennessee Valley Authority, instead of paying once again to restore the system that failed.