A faulty price feed triggered the abnormal liquidation of 47 borrowing positions holding roughly 3 million USD in collateral on Vesu, a lending protocol built on Starknet, according to an incident disclosure published September 5 — and the episode is a textbook reminder that in DeFi, your loan is only as reliable as the data feed that watches it.
By David Chen | September 5, 2026
The liquidations occurred between 04:08 and 04:10 UTC on September 4, after an upstream price source operated by Pragma supplied incorrect data to several Vesu liquidity pools. Within two minutes, automated liquidation bots removed approximately 3 million USD in collateral before the feed corrected itself.
The Hook: Two Minutes of Bad Data, 47 Wiped Positions
Vesu explained that during the brief window, the faulty prices made 47 borrowing positions appear eligible for liquidation across several pools. Liquidators — automated programs that close undercollateralized loans in exchange for a bonus — acted on the bad numbers and swept the collateral before the feed returned to correct values.
- 47 positions liquidated across several Vesu pools during a two-minute failure window.
- 3 million USD in collateral removed before the Pragma feed self-corrected.
- No protocol vulnerability — Vesu says its contracts were “operating as designed.”
- Recovery talks underway with Pragma, StarkWare, the Starknet Foundation, and pool curators.
How a Price Feed Can Liquidate a Healthy Loan
In DeFi lending, you deposit assets worth more than your loan — like leaving a 2,000 USD watch as security for a 1,000 USD loan. The protocol constantly checks the value of your collateral using an oracle, an outside service that collects market prices and publishes them on-chain, because smart contracts cannot read market data by themselves.
If the oracle suddenly reports that your collateral is worth far less than it really is, the protocol believes your loan has become undercollateralized and opens it to liquidation. Your position was perfectly healthy by real market prices — but the contract only knows what it is told. That is exactly what happened on Vesu: bad inputs, not faulty execution. An oracle normally handles data sourcing, aggregation, and on-chain delivery, and a failure at any of those three stages can pass a wrong value to an otherwise functioning contract.
The Core Conflict: Not a Hack, but Not Harmless Either
Vesu was quick to separate the incident from a smart contract exploit. The protocol said its contracts contained no vulnerability and it has no patch to deploy, because the liquidation engine responded exactly as programmed to the prices it received. Pragma has since worked with the relevant organizations to deploy a fix addressing the source of the error, and liquidity pool curators suspended the affected pools as a precaution until the fix is reviewed.
That distinction offers little comfort to the 47 users whose collateral was seized. Blockchain transactions are generally final once confirmed, so there is no automatic reversal. Any recovery requires voluntary returns from liquidators, protocol-controlled funds, or another compensation arrangement agreed by the parties — and Vesu has not said which route it plans to use, how much of the 3 million USD is recoverable, or whether a reimbursement timeline exists.
What This Means for You
If you used Vesu’s Earn product, the protocol’s guidance is direct: keep your position open. Closing an Earn position before the recovery process completes may disqualify you from a refund. Borrowers who were liquidated during the two-minute window were asked to open a support ticket through Vesu’s Discord, with wallet addresses and transaction details helping identify affected positions on-chain.
The bigger lesson applies to every DeFi user: oracle risk is invisible until it isn’t. Unlike a bank account, a permissionless lending protocol comes with no government-backed insurance — the FDIC protection that covers eligible bank deposits does not extend to DeFi positions. A comparable event hit Aave in March 2026, when a stale parameter caused an estimated 26 to 27 million USD in unintended wstETH liquidations, after which Aave examined oracle update rates and fallback systems. Vesu has announced no comparable changes yet; Pragma’s root-cause fix is the only technical measure confirmed so far.
The Verdict
Vesu’s contracts did what they were told, liquidators did what they are built to do, and 47 users paid the price for a two-minute data glitch. The incident is a reminder that in DeFi, “the code worked” and “the outcome was fair” are two very different statements. Vesu says a full technical report is coming, and the recovery effort — involving Pragma, StarkWare, and the Starknet Foundation, part of the ecosystem behind Starknet’s STRK20 privacy rollout that Vesu supports — will show how seriously the ecosystem treats users caught in machinery working as designed. For market context, Bitcoin traded around 79,725 USD, Ethereum near 2,458 USD, and Solana around 103 USD at the time of this snapshot, according to CoinGecko data.
The cryptocurrency market remains highly volatile. This article is for informational purposes only and does not constitute financial advice.
Disclaimer: This article is for informational purposes only and does not constitute financial advice.
04:08 to 04:10 utc, bots racing a bad feed while humans slept. whoever ran those liquidators knew exactly what they were doing
pragma corrected itself in 2 minutes and vesu is still doing incident theater instead of refunds. 3m is marketing budget money for a lending app, pay the users
a 3m refund also buys them the median plus staleness fix shipping without user pressure. cheap insurance on both ends honestly
two minutes of bad pragma data and 3 million gone. liquidator bots dont care if the price is real, they just race the correction
^ exactly. vesu saying the contracts operated as designed makes it worse imo, the design trusts one feed with no sanity check
a two source median with staleness cutoff wouldve killed this instantly. primitives were fine, the config wasnt
two source median plus a staleness cutoff is lending 101. vesu shipping without either is wild, configs are where protocols actually die
median or not, a 2 minute bad feed still gets raced before any fix triggers. staleness cutoff plus deviation cap together is the only combo that catches both failure modes, one alone leaves a gap
correct take in this whole thread. either fix alone still gets raced inside the 2 minute window, you need both to catch a bad median and a stale one
a 2 min bad feed at 4am utc on a thursday. whoever picked that window to run the exploit-ish liquidations knew exactly what they were doing
one of the 47 here. feed glitched for 120 seconds at 4am and my healthy position was gone before i woke up. bots dont check if the price is real
the worst part is the liquidators profited off a feed nobody chose to trust. vesu comping all 47 positions is the minimum acceptable outcome, not a generosity play
this should be pinned. a 4am liquidation from one bad pragma feed and the fix is a median, a thing that should have shipped day one
one of the 47 as well? did vesu reach out to you directly or is the recovery talk still just starkware internal whispers. hope the 3M finds its way back
did you get made whole or still waiting? everyone mentions the 3m refund, nobody has posted a timeline yet
still waiting here. vesu says refund is coming but 04:08 utc victims have zero tx hashes to check. publishing the payout list would take them five minutes
47 positions gone in the time it takes to make coffee. hope the refund actually lands, you were the collateral in someone elses config decision
47 positions wiped that were probably healthy. hope the starkware and starknet foundation recovery talks actually go somewhere for those users
starkware refunding would be smart, 3m is cheap compared to the trust vesu loses on starknet if they dont
two minutes. 47 positions gone because one upstream pragma feed glitched at 4am. oracle risk is still the most underrated way to get liquidated in defi
^ this is why i only borrow on chainlink-fed venues. vesu users found out the hard way what a single feed dependency costs
chainlink helps but even then you want a deviation cap. one source, no sanity band, 47 positions, thats just lazy risk design
3M wiped in 2 minutes and the post mortem reads like a shrug emoji. starknet lending deserves better oracles than this ngl
everyone hyping starknet security and then a bad pragma feed nukes 3M in collateral lol. l2s are great until the oracle isnt
two source median with a deviation cap gets suggested after literally every oracle incident since bzx in 2020. vesu will ship it next month and call it innovation lol
Publishing the oracle config would let users see the single feed dependency before depositing. Nobody audits the config until it liquidates them.