"article": "In October 2024, Bill Dudley wrote a sentence that should have repriced every regional bank in the United States. It did not.\n\nDudley ran the Federal Reserve Bank of New York from 2009 to 2018. He sat in the room where the choice gets made between liquidity and last rites. When a man with that résumé says the resolution regime is too weak to absorb a bank failure, you are not reading an opinion column. You are reading a bug report filed by someone who had root access.\n\nThe bug report was specific. Capital requirements are being eased. The machinery that is supposed to unwind a failed bank when the capital runs out has not been strengthened to match. Two halves of one system, moving in opposite directions at the same time.\n\nThe market's response was silence.\n\nSilence is the only honest consensus mechanism, and it is usually wrong. When nobody reprices on a statement like that, it does not mean the statement was false. It means the statement was classified as noise by people who have never read an assembly listing in their lives.\n\nThe code whispered what the pitch deck screamed. Nobody was listening to the code.\n\nI have spent nine years reading systems where the front end said one thing and the back end said another. I audited an ICO whitepaper at sixteen and watched it rug six months later. I found an integer overflow in a Compound governance upgrade that could have drained fifty million dollars. I pulled two hundred terabytes of transaction logs from a collapsed exchange's multi-signature structure and found commingling where the public had been promised segregation. This article is about a traditional banking regulation story that has, on its surface, nothing to do with crypto. It has everything to do with crypto. The architecture Dudley is describing is the same architecture that failed on-chain in 2020, 2022, and 2023 — and the crypto industry has been shipping the broken version of it for a decade while calling it innovation.\n\n## The Two Numbers That Define a Banking System\n\nStart with the definitions, because the entire policy debate hides inside them.\n\nA capital requirement is a ratio. It says: for every dollar of risk-weighted assets you hold, you must fund some fraction of it with equity rather than debt. Call it a collateral ratio. It is a number. It is measurable. It is reported quarterly. It is auditable in principle. Regulators can look at it and say, with reasonable confidence, whether a bank is solvent on a given Tuesday.\n\nA resolution regime is not a number. It is a promise.\n\nA resolution regime is the set of legal and operational procedures a government uses to take over, stabilize, sell, or liquidate a failing bank without triggering a system-wide panic — without pulling the plug on the payment system, without freezing every depositor, without forcing the taxpayer to write the check in the middle of the night. In the United States, the primary tools are the FDIC's receivership process for ordinary banks and Title II of the Dodd-Frank Act — the Orderly Liquidation Authority — for the enormous ones whose failure would otherwise cascade.\n\nRead those two definitions side by side and the problem becomes structural rather than political.\n\nCapital requirements can be modeled. Resolution regimes can only be rehearsed. One is arithmetic. The other is a narrative about what will happen when the arithmetic fails, performed by institutions that have every incentive to be wrong about it on purpose.\n\nDudley is not arguing against easing capital requirements. He is arguing that easing them without strengthening resolution is a one-sided trade. You are reducing the buffer and refusing to upgrade the parachute. That is not deregulation. That is a configuration error, committed with full knowledge, at the level of a system too big to fail and too proud to say so.\n\nI have seen this exact error before. It has a name in my field. It is called a protocol with a liquidation engine that exists only in the documentation.\n\n## The Parachute That Was Never Packed\n\nTo understand what Dudley is pointing at, you have to go back to March 2023, because March 2023 was the live-fire test of the resolution regime, and the regime failed it in public.\n\nSilicon Valley Bank held roughly $209 billion in assets when it failed on March 10, 2023. Signature Bank held roughly $110 billion. First Republic held roughly $229 billion when it went down on May 1. Three of the four largest bank failures in American history happened inside eight weeks. The word \"systemic\" stopped being a word and became a signature on a piece of paper.\n\nNow watch the specifics, because the specifics are where the architecture leaks.\n\nSVB's problem was not a mystery. It held a securities book full of long-duration assets bought during the zero-rate era. When the Federal Reserve hiked, the market value of that book fell below its book value, and the difference was sitting in the held-to-maturity bucket where accounting rules allowed the bank to pretend the loss did not exist. The unrealized loss on those securities was, by most estimates, north of fifteen billion dollars. That is not a rounding error. It is the gap between solvency and insolvency, parked off the income statement.\n\nThen the depositor base. SVB's deposit base was overwhelmingly uninsured — the overwhelming majority of its deposits sat above the $250,000 FDIC insurance limit, because its depositors were venture funds and startups whose operating cash exceeded that number as a matter of course. On paper, the bank was fine. In practice, it was a single social-media rumor away from a bank run, because every depositor had a rational incentive to be first out the door.\n\nOn March 10, the run happened. By the weekend, the FDIC was in receivership. By Sunday, the government had invoked the systemic risk exception to make uninsured depositors whole anyway — a decision that was technically legal and philosophically enormous, because it quietly acknowledged that the resolution regime could not, on its own, contain the failure without breaking a promise the system had spent a decade pretending it would keep.\n\nHere is the part that matters for Dudley's argument. A resolution regime is only as strong as the moment it is invoked, and it is invoked at the precise moment when the people running it are least able to think clearly. You cannot resolve a bank in a calm afternoon. You resolve a bank on a Sunday night, with the futures market already open in Asia, with a board that has spent the last week lying to you, with a deposit base that has already decided to leave, and with a legal framework that was written for a world where depositors moved money at the speed of a wire instead of the speed of a tweet.\n\nThe 2018 rollback of the Dodd-Frank SIFI threshold — the move that lifted the \"systemically important\" designation from institutions between $50 billion and $250 billion — is the direct ancestor of the March 2023 failure. SVB sat under that raised threshold. It was not treated as systemically important by the framework. It turned out to be systemically important in practice. Truth hides in the assembly, not in the press release, and the assembly said the threshold was designed for a world that no longer existed.\n\nSo when Dudley, in October 2024, says that easing capital requirements without strengthening resolution is dangerous, he is not theorizing. He is pointing at a smoking crater and saying: the next one is already scheduled.\n\n## What Crypto Already Proved About Liquidation Engines\n\nThe banking world is now learning something the crypto world learned the hard way, in public, with a smaller balance sheet and no central bank to call.\n\nA bank's resolution regime is a liquidation engine. It is the code that runs when the collateral ratio is breached. It is the thing that is supposed to convert a failing institution's assets into something that satisfies its liabilities without setting the rest of the system on fire.\n\nIn crypto, we have built liquidation engines over and over. Some of them work. Most of them do not. And the failure modes are remarkably consistent, because the failure mode is always the same: the engine assumes there will be a buyer, and at the exact moment the engine runs, there is no buyer.\n\nTake March 12, 2020. Ethereum fell roughly forty-five percent in a single day. On-chain, MakerDAO — the protocol that issues DAI against collateral — ran its liquidation auctions. The auctions were supposed to attract bidders who would purchase the seized collateral at a discount and, in doing so, repay the debt. The bidders did not come. The network was congested. Transaction fees exploded. The auction mechanism, designed for a world where someone would always outbid zero, received bids of zero. The protocol ate the loss. Roughly eight million dollars of bad debt was socialized, and MKR — the governance token — was minted and sold to cover it.\n\nThat is a resolution regime. It ran. It failed. And the failure was not a market failure primarily. It was an architectural failure. The engine assumed a liquidity condition that the engine itself was responsible for destroying.\n\nI have audited versions of this pattern more times than I can count. In 2020, while I was still a student, I spent two weeks inside the Compound Finance governance contract and found a subtle integer overflow in a newly proposed upgrade. That bug, if it had shipped, could have drained somewhere in the neighborhood of fifty million dollars. I reported it privately to the core developers through a secure channel. They patched it within forty-eight hours. Nobody wrote an article about it. Nobody minted a governance token to celebrate it. The most important security work in this industry is silent, and that is not a bug — it is a feature. You do not hear about the failures that did not happen. Which is precisely why every exploit that does happen takes on an outsized narrative weight. Every exploit is a story poorly told. The market remembers the story, not the architecture.\n\nNow map that back onto banks.\n\nWhen the FDIC resolves a failed bank, it is running an auction. It is trying to find a buyer for the assets — a healthy institution willing to take on the deposits and the loan book. When there are buyers, resolution works. When there are no buyers, because the entire sector is under stress at the same time, resolution becomes nationalization by another name. That is what happened in 2008. That is what happened, in miniature, in March 2023, when the government made depositors whole across the board because no private buyer existed for the combination of assets and liabilities at any price the system was willing to print.\n\nThe resolution regime is not a mechanism. It is an option on a buyer, and the option is priced at zero in every scenario where it needs to be exercised.\n\n## Capital Requirements as a Collateral Ratio, and Why the Ratio Lies\n\nHere is where the crypto analogy stops being a metaphor and starts being a diagnosis.\n\nIn DeFi, a lending protocol like Aave or Compound requires borrowers to over-collateralize. You deposit ETH, you borrow against it, and the protocol watches the health factor. When the health factor drops below one, the protocol liquidates you. The over-collateralization is the buffer. The liquidation engine is the resolution regime. The oracle is the reporting layer.\n\nEvery experienced auditor knows where the real risk lives. It is not in the collateral ratio. It is in the oracle.\n\nIf the oracle reports a price that is fifteen seconds stale, the engine liquidates at the wrong time. If the oracle is manipulable, the engine liquidates for the wrong reasons. If the oracle is down, the engine does not run at all, and the bad debt accumulates invisibly until the next block. The collateral ratio — the number — is only as good as the data feeding it, and the data is only as good as the mechanism delivering it.\n\nNow look at a bank.\n\nThe bank's risk-weighted assets are computed using models. Those models are the oracle. They report the value of the loan book, the duration risk of the securities book, the default probability of the counterparty exposure. The models are internal. They are audited, in principle, by the same firms that signed off on SVB's books the week before it failed. They are calibrated, in practice, by the desire not to be the bank that reported a capital shortfall and triggered a run.\n\nThe capital requirement is the collateral ratio. The model is the oracle. And the oracle is frequently wrong in exactly the direction that makes the ratio look compliant.\n\nI audited a set of NFT projects in 2021 for a potential investment fund. Fifty collections. I was drawn to one set of generative art algorithms — genuinely beautiful work, mathematically elegant, the kind of code you read twice because the first read is aesthetic and the second read is arithmetic. Then I opened the royalty logic. There was a proxy pattern, buried three layers deep, that allowed royalty evasion. The creators would not be paid on secondary sales. The art was real. The contract betrayed it anyway.\n\nBeauty is the most sophisticated rug pull. It is the layer that makes you stop auditing. And a capital ratio is beautiful in exactly the same way. It is a clean number. It is a single decimal. It is the kind of number that disappears an investor's anxiety. Which is precisely why it is the number that gets manipulated.\n\n## The Moral Hazard Engine\n\nDudley's argument comes down to a mechanism that is worth naming precisely, because the name is the diagnosis.\n\nWhen you ease capital requirements, you lower the cost of holding risk. Capital is expensive. Equity is more expensive than debt. A bank that must fund a larger share of its assets with equity earns less on each dollar of assets. A bank that can fund more of its assets with debt earns more, and takes more risk, and — critically — has less of its own money on the line if the risk goes bad.\n\nWhen you fail to strengthen resolution at the same time, you send a second signal. You signal that if the risk goes bad anyway, the failure will be handled by a regime that is known to be weak, which means it will be handled by the mechanism of last resort — which means it will be handled by the taxpayer, or by the depositor, or by the system's willingness to invoke the systemic risk exception one more time.\n\nStack those two signals. Lower cost of risk. Softer landing if risk turns bad.\n\nThat is not a policy. That is a subsidy on leverage with an implied put option on failure. In the language of the industry I work in, it is a protocol that pays you to borrow against yourself and guarantees you will not be liquidated if you go underwater.\n\nWe built that protocol on-chain too. It was called Terra. It was called Celsius. It was called FTX. In every case, the marketing described a system with risk management. The architecture described a system with a promise that the risk would be contained by people who had every incentive to hide it.\n\nIn 2022, I sat in Toronto and audited the multi-signature wallet structure of a collapsed exchange. Two hundred terabytes of transaction logs. I did not give interviews. I did not post threads. I went through the data and found what the public claims had said could not exist: funds that were represented as segregated, moved in ways that made clear they were not. The commingling was not a bug in the code. The code executed exactly as written. The commingling was a bug in the story the code was wrapped in.\n\nI filed the report. It was emotionless. It was formatted like a legal brief. It has no adjectives that imply feeling in it, because feeling does not survive cross-examination. That period taught me something that applies directly to the Dudley position: in a crisis, silence and precision beat volume every time. The people making the most noise are usually the ones who have not read the assembly.\n\n## Stablecoins Are Bank Deposits With Worse Oracle Design\n\nThere is a version of this story that the crypto industry cannot pretend is external. It is the stablecoin.\n\nA stablecoin is a bank deposit with a stranger's balance sheet behind it. When you hold USDC, you hold a claim on a reserve pool that is supposed to be equivalent to a dollar. The reserve pool is the collateral. The issuer is the bank. The attestation is the oracle. And the redemption mechanism is the resolution regime.\n\nIn March 2023, when SVB failed, Circle had roughly $3.3 billion of its USDC reserves parked at that institution. USDC depegged to just above eighty-seven cents. The depeg was not caused by a run on USDC primarily. It was caused by the market's sudden, correct recognition that the reserve attestation was lagging the reserve reality by an amount that mattered. The oracle was stale precisely when freshness was the whole game.\n\nLook at the architecture. An attestation is not an audit. An audit is not a guarantee. And a guarantee, in a system with no bankruptcy waterfall and no resolution regime, is a wish.\n\nCircle's reserves were, in the end, fine. The peg recovered. But the recovery was not achieved by the issuer's resolution mechanism, because the issuer had no resolution mechanism. The recovery was achieved by the federal government's willingness to backstop the depositors of SVB through the systemic risk exception — the same mechanism Dudley is now telling us is too weak to be relied upon.\n\nThe stablecoin market's solvency was outsourced to the bank resolution regime. And the bank resolution regime is, by Dudley's own account, not strong enough to carry that weight.\n\nSit with that for a moment. The largest dollar-denominated instruments in crypto are structurally dependent on a resolution regime that a former central banker says is underbuilt. The crypto industry has spent years arguing that it does not need banks. The architecture has been quietly arguing the opposite the entire time.\n\n## The Inversion Nobody Wants to Name\n\nHere is the contrarian angle, and it is the one that both the crypto natives and the banking traditionalists get wrong.\n\nThe standard crypto position is that banks are obsolete — slow, fractional, levered, and politically protected. The standard banking position is that crypto is a casino with no consumer protection and no systemic relevance. Both positions are comfortable. Both are wrong. And they are wrong in the same direction, because the actual movement in the last several years has been banks adopting crypto architecture and crypto adopting bank architecture, with neither side admitting it.\n\nTokenized deposits are the clearest example. When a major bank issues a token representing a claim on its balance sheet and settles it on a distributed ledger, it is not adopting crypto's philosophy. It is adopting crypto's plumbing. The balance sheet stays the same. The leverage stays the same. The resolution regime — the thing that determines what happens when the balance sheet fails — stays exactly where it was, which is to say, in the hands of a government that Dudley says is not prepared.\n\nSame mechanics from the other direction. When a lending protocol structures a real-world-asset vault backed by Treasury bills, it is not escaping the banking system. It is re-importing the banking system's duration risk, its interest rate sensitivity, and its dependency on a custodian, and wrapping it in a smart contract that has fewer bankruptcy protections than the underlying instrument ever had.\n\nWhat the bulls got right — and I want to be precise about this, because a teardown that does not acknowledge the strongest counterargument is a press release in reverse — is that these are genuinely new architectures for settlement. Distributed ledgers do not just move the same money faster. They change who can verify what, at what cost, and against whose permission. That change is real. Interbank settlement that takes a day and a half and requires a correspondent chain is a genuinely fragile system. Replacing the correspondent chain with a cryptographically verifiable layer is a structural improvement, not a marketing claim.\n\nBut the bullish case for tokenization almost never prices the resolution question. It prices the speed. It prices the composability. It prices the twenty-four-hour market and the atomic settlement and the programmable collateral. It does not price the fact that when the tokenized deposit system has its first real failure — a custodian that goes down while its token stays live, a vault that cannot redeem because the underlying asset is frozen — the resolution regime that has to catch it will be the same underbuilt regime Dudley just described. And the tokenized deposit will inherit every flaw of the bank behind it, plus one: the on-chain layer that makes recovery harder, not easier.\n\nTokenization does not remove the resolution problem. It relocates it, and it makes it harder to see.\n\n## Where the Mechan
