How price manipulation attacks drain DeFi lending protocols

DeFi lending protocols lost hundreds of millions of dollars this year to a single, cheap trick. Blockchain intelligence firm TRM Labs counted 32 price manipulation attacks in 2026, the most it has ever recorded in a single year. No stolen keys. No secret backdoor.

TRM Labs published that figure on 31 August 2026. A day earlier, an attacker had borrowed an estimated $75 million out of Tectonic, the largest lending market on the Cronos chain, using a token almost nobody was trading. Price manipulation now makes up about one in eight crypto hacks, the firm said, compared with roughly one in 17 in 2022.

What a price oracle does

A lending protocol has to know what your deposit is worth before it lets you borrow against it. Here is the catch. A smart contract cannot look up a market price on its own. It lives on the blockchain and sees only what someone puts in front of it.

That job falls to an oracle. An oracle is software that reads a price from somewhere, an exchange, a trading pool, a data provider, and writes it onto the chain where the contract can use it. Deposit a token, and the oracle tells the protocol what it is worth. Ask to borrow, and the oracle sets your limit.

So the whole system rests on one assumption: that the price the oracle reports is real. Break that assumption and everything built on top of it pays out wrong. That is the door attackers walk through.

How the attack works

Most of these attacks run the same play. Find a token with barely any trading behind it. Pump its price with a wave of buying, often bankrolled by a flash loan, a loan taken and repaid inside one transaction that needs no collateral of its own. Post the inflated token as collateral. Borrow as much real money against it as the protocol will hand over. Leave before the price drops back to earth.

Flash loans are what make it cheap. An attacker needs no war chest. They can borrow millions of dollars for a few seconds, use the money to shove a thin market around, and repay it in the same transaction, all before the block closes. If any step fails, the whole thing unwinds and they lose only the network fee. That asymmetry is the point. Small risk, large prize.

Tectonic shows the shape of it. The attacker targeted TONIC, the protocol's own governance token. In the week before the attack, TONIC had traded about $305,000 in total, less than a mid-size car dealership turns over in a month. The attacker bought aggressively and drove the price up roughly 100 times in about 20 minutes, according to TRM Labs. That inflated valuation became collateral for a loan worth tens of millions of dollars. Then it was gone.

How big was the loss? Even that is not settled. TRM Labs put it near $75 million. One separate analysis put it at $119.5 million. The two figures have not been reconciled, and the person behind the attack has not been named.

Why the same trick keeps working

None of this is new. Back in 2022, an attacker inflated the token MNGO and borrowed against the fake value, pulling roughly $117 million out of Mango Markets, one of the largest cases on record, security firm CertiK has documented. What changed in 2026 is the sheer count.

Almost always, the weak point is the price source. Some protocols read a price from a single on-chain pool, so whatever that pool shows in a given block gets treated as gospel. Move the pool, move the price. Others use a time-weighted average, which smooths out brief spikes, though a patient attacker can hold a bent price across the whole averaging window. Protocols that pull from deep, off-chain feeds such as Chainlink are harder to fool, because no single trade shifts the reference. Newer projects often reach for the simple on-chain option anyway, because a fresh token may have no deep feed to point at, and building one costs time the team would rather spend shipping.

Lending protocols do carry a safety valve. They let you borrow only a fraction of your collateral's value, and they liquidate positions that slip underwater. Both defenses assume the reported price is honest. When the price itself is the lie, that fraction is measured against a fake number, and the liquidation fires too late, after the borrowed money has already left.

This is not a hack in the way most people imagine one. The contract usually does exactly what it was written to do. It just trusts a number it should not. That is why a clean code audit does not save a protocol here, and why the same attack keeps landing on new victims.

When the price feed itself is the target

Not every case is a spot-price pump. Sometimes the attacker goes after the plumbing that delivers prices at all.

In July, someone drained about $18 million from Ostium, a trading protocol, by feeding its price system reports stamped with future dates. Those bad timestamps made losing trades read as winners, and the vault paid out, CoinDesk reported. Security firm Blockaid flagged the transaction on 15 July 2026. Different method, same result: wrong data in, real money out.

One number frames the wider problem. Across the first half of 2026, TRM Labs logged 207 crypto security incidents and about $972 million stolen in total. Price manipulation is only one slice of that, but it is the slice growing fastest.

Someone always eats the loss, and it is rarely the people who built the protocol. When a lending pool is drained, the hole comes out of the funds other users deposited to earn a return. Some protocols keep an insurance reserve for exactly this moment. Many do not, or the reserve is far too small to cover a nine-figure gap. Depositors can log in to find part of their balance simply gone.

What to watch

Fixes are not a mystery, and that is the frustrating part. Deeper price feeds. Tighter rules on which tokens can back a loan. Caps on how much can be borrowed against a thin asset. The tools exist. Using them costs a protocol the very thing that pulls users in, which is a long menu of tokens you can borrow against.

Watch where the next attacks land. If they keep hitting small, permissionless markets that list almost any token, the pain stays with people chasing yield on obscure coins. If they start reaching larger, mainstream lending pools, the story gets serious. TRM Labs offers one steadying number here. The median loss per incident this year is about $219,000. The frightening totals come from a small group of outliers, not the everyday case.