Which stablecoins have lost their peg, and what happened to the money?

Over the weekend of 11 March 2023 a USDC token traded at 86 cents.

By the Monday it traded at a dollar again. In between, a US bank failed on the Friday, three federal agencies said on the Sunday evening that its depositors would get all of their money back on the Monday, and Circle began processing redemptions again that Monday morning.

That is the shape of a depeg, and it is not the shape most merchants picture. The price on an exchange is one number. What sits behind the token, and whether anybody is standing at the redemption window, are two others. They come apart under stress, and that weekend is where anyone can watch them do it.

What follows is the record with dates on it: the cause, how long it held, and what a merchant holding the token could do while it did. What backs each design is a separate question, answered on a separate page. This one is the failure history.

What does a depeg measure?

A price on a secondary market, and nothing beyond it.

Stablecoins are traded on exchanges in a way bank deposits are not, and the exchange price is whatever sellers will take at that minute. In a rush, the sellers are the people who cannot wait. Redemption at the issuer is a separate door, and it keeps its own hours.

March 2023 pulled the two apart in public. A Federal Reserve staff note on the episode puts the selling pressure on the closed door rather than on missing money: suspension of primary-market redemptions over the weekend "contributed to selling pressure on USDC in the secondary markets", and the price "fully recovered once Circle began processing redemptions on Monday, March 13".

So 86 cents was the price of getting out on a Saturday. A merchant who took it turned a banking-hours problem into a realised loss; a merchant who read the news on Tuesday never had one. Neither sentence is a rule for the next event. Both are the arithmetic of a single weekend, which happens to be the one with the most paperwork behind it.

Which pegs have broken so far?

Seven. Hardly a repeated cause among them.

The table at the foot of this article is the list, and its middle column is the finding. TerraUSD is the one everybody remembers, and the SEC's account of it is flat enough to quote whole — in May 2022 UST "depegged from the U.S. dollar, and the price of it and its sister tokens plummeted to close to zero". The thing holding its price up was the thing the same panic was selling.

Nothing else in the table failed that way. One fell because a bank holding its reserves did. One swapped good collateral for bad inside the module built to hold its peg steady, and two more went down in the same exit, each to a different low. A fifth is a reserve claim that had been untrue for most of two years and stayed private for three more. The last two are not price events at all — a regulator's instruction, and one exchange's index.

Then read the same seven the way an alert would. Five of them moved a price somewhere a merchant could have watched. The other two never moved one, which is why a price watch could not have seen them at any threshold. A depeg is the visible subset of the problem, and it is a poor thing to name a risk control after.

Why did one weekend produce four different prices?

Because each token's low was set by its own plumbing.

USDC fell furthest because Circle could not withdraw $3.3 billion of reserves from Silicon Valley Bank, around 8% of the total at the time. Dai fell for a reason of its own making: its peg stability modules let holders swap in, and the Federal Reserve note describes the result as a re-balancing of Dai's collateral pool "from 'higher-quality' assets such as Ethereum, over-collateralized at 150%, towards distressed USDC, under-collateralized at less than 100%". Pax Dollar lost over 400 million tokens out of its module over the same period — more than half its outstanding supply — and bottomed around 91 cents. Gemini Dollar reached about 96 cents.

Two tokens went the other way. The same note records a flight to safety into USDT and BUSD, which "appreciated over the weekend and traded at a price marginally above their one-dollar peg".

Hold those five names next to their labels and the label stops predicting anything. USDC, Pax Dollar, Gemini Dollar, USDT and BUSD were all fiat-backed that weekend. One of them traded at 86 cents; two of them traded above a dollar. The category a token belongs to told a merchant nothing that weekend, and the name of its bank told them the whole story. Dai is the instructive one: its collateral was on-chain and legible to anybody who cared to look, and it imported the problem anyway, through the very machinery meant to keep it out.

Can a steady price hide an unsteady reserve?

For most of two years it did, and nobody watching the price knew.

On 15 October 2021 the CFTC fined Tether $41 million. The order's finding is specific about how long the gap ran: Tether held enough fiat reserves to back the USDT in circulation "for only 27.6% of the days in a 26-month sample time period from 2016 through 2018", and the reserves "were not 'fully-backed' the majority of the time".

Three years separate the conduct from the public record of it. Whatever a merchant checked in 2017 before deciding to hold USDT, it was not this — this did not exist yet in a form anyone outside the company could read. Which is the honest version of "monitor your stablecoin risk": the number you can refresh every minute is not the number the failure lives in.

What ends a stablecoin without a depeg?

An instruction to the issuer.

Paxos published one on 13 February 2023. "Effective February 21, Paxos will cease issuance of new BUSD tokens as directed by and working in close coordination with the New York Department of Financial Services (NYDFS)." The same statement kept the exit open: BUSD "will remain fully supported by Paxos and redeemable to onboarded customers through at least February 2024", in dollars or converted into Pax Dollar.

The peg is not what moved. What the token acquired was an end date, roughly a year out, announced in a press release rather than on a price feed. A treasury holding BUSD had plenty of time and no automatic prompt to use it — and a year is exactly long enough for a thing like that to be forgotten twice.

Was that the asset, or the venue?

In October 2025, for forty minutes, it was the venue.

Binance's own announcement fixes the window at 21

to 22 UTC on 10 October 2025 and covers three tokens at once: USDE, BNSOL and WBETH. It promises affected users automatic compensation within 72 hours, at the difference between their liquidation price and the market price at 00 UTC the next day.

The remedy is where this one gets interesting. Binance folded the redemption price into the index weights for all three tokens and added a minimum price threshold to the USDE index rule.

An issuer with a reserve problem repairs the reserve; an exchange that starts weighting redemption into its index is repairing a price it published. So a figure quoted from that night carries a question with it. Which order book, and was your money anywhere near it?

What does a depeg do to a Paymos balance?

It moves the value and leaves the position exactly where it was.

A balance here is held per asset rather than kept as a running dollar figure, and the same asset arriving over several networks pools into that one balance; the merchant picks the exit network at the point of creating a payout.

Three of the platform's five settlement assets are in the record above — USDC, Dai and USDT.

Paymos settles in the asset the payer sent. Nothing turns it into dollars, BTC or ETH, no bank account sits at the other end of a payout, and nothing is ever scheduled or triggered automatically. So no balance leaves an asset on its own. The position a merchant wakes up holding is the position it went to bed with.

The leaving is fast, once somebody decides. There is no queue on this side of it: the request becomes a transaction. After that, the destination network decides when it lands.

A pair of limits is better learned in advance than on the day. An asset-and-network payout route can be unavailable for a while, and it is refused when the payout is created rather than parked in a queue; the balance is untouched and another destination network works. Payouts also reach 11 of the 13 networks: NEAR and Sui take payments and are not payout destinations. The balance is not put to work while it waits. Nobody lends against it and nobody earns on it.

Which levers does a merchant have?

Two, and neither of them is an alert.

The first is pulled at invoice creation rather than in project settings. An amount priced in a fiat currency leaves both the coin and the network to the payer. Naming a crypto symbol narrows the checkout to that asset across every network the project accepts. Naming a symbol together with a network leaves nothing to choose, and the payer lands straight on the address. Whichever of the three a merchant picks, it has settled what its balance will be denominated in long before anybody looks at a price.

One mechanic looks like a third lever and is not — the rate lock. An amount entered in a fiat currency is converted at a rate locked when the payer confirms the asset and network, before any transfer is sent, so the token figure is settled and on screen before a wallet opens. It closes the gap between quoting a price and receiving it. It fixes how many tokens arrive, and it has nothing to say about what those tokens are worth a week later.

The second lever is the exit, and the section above is all of it. One thing to add about what you are exiting: the five settlement assets are not five versions of one risk. Four of them track a dollar. XAUT is gold-backed and follows the gold price. Changing asset changes which exposure you are carrying, not whether you carry one.

None of that is protection. The price is the price — nothing on this side moves against it and no loss gets made up.

rewards is duller than any monitoring product: knowing which asset the balance is in, and being in a position to decide to move it on a Saturday.