Moonwell's $8.7M Loss and How Olympix Would Have Prevented It
Date: August 27, 2026
Chain: Base
Loss: about $8.7M bridged out by the attacker, with roughly $9.13M of borrower obligations left behind in Moonwell's markets
Class: a collateral exchange rate inflated by tokens sent straight to the market contract, past a supply cap that only watched deposits
Between 06:09 and 09:30 UTC on August 27, 2026, an attacker turned roughly $7.5M of a thinly traded token into $11.03M of borrowed cbBTC, WETH, USDC and wstETH on Moonwell's Base markets. The token was MAMO, and Moonwell capped its MAMO market at 20 million tokens of supply. The attacker stayed under that cap on paper while putting about 68 million MAMO into the market contract, because 53.4 million of them never went through the deposit function the cap was attached to. They were simply transferred in.
Liquidators reacted fast and seized nearly all of the attacker's collateral, but by then the borrowed assets were already being swapped and bridged off Base. Moonwell's own post-mortem, dated August 28, put the borrower obligations still outstanding from the incident at about $9.13M.
Three Numbers That Price a MAMO Position
Supplying MAMO to Moonwell mints mMAMO, a receipt token for a share of the market. How much a position can borrow depends on three numbers multiplied together, as QuillAudits' analysis lays out: the mMAMO balance, the exchange rate that says how much MAMO backs each mMAMO, and the oracle price of MAMO. Moonwell then lends up to the market's collateral factor, 50% for MAMO, against that value.
The exchange rate is computed from the MAMO the contract holds against the mMAMO in circulation. Supply through the front door and both sides grow together, so the rate stays put. Send MAMO to the contract directly and only the backing grows, so every existing mMAMO becomes worth more MAMO, and nobody minted anything. The 20 million MAMO supply cap was checked on the deposit path, so a direct transfer never touched it.
The oracle side had its own weakness. According to QuillAudits, MAMO's price came from a spot-derived feed the attacker could move by trading into the same thin liquidity. Two of the three numbers were reachable by the borrower, and the attack used both.
Weeks of Buying, Then a Morning of Borrowing
The money came in through Tornado Cash. A linked account received 800.098 ETH in several withdrawals, sold it through CoW Swap, and bridged the proceeds to Base through Circle's CCTP. The attacker then spent about $7.50M buying roughly 94.3M MAMO across dozens of trades on thin DEX liquidity.
On August 27, the attacker supplied 15,089,595 MAMO to Moonwell the normal way, worth about $159,904 at the reference price of $0.010597. Then came two direct transfers to the mMAMO contract, 34,910,396.68 MAMO and 18,482,893.63 MAMO, for 53,393,290 MAMO in total. No mint event accompanied either one. The exchange rate went from about 0.020513 to 0.075460, a jump of roughly 3.68x, and because the attacker held about three quarters of all mMAMO, most of that new backing landed on their own position.
Next, the attacker bought more MAMO into the same thin pools. Per Moonwell's post-mortem as reported by CoinStats, the source feed climbed from about $0.0106 to a peak of $0.43127363, roughly fortyfold, and the highest price Moonwell accepted as valid was $0.40248571. With a bigger exchange rate and a fortyfold price stacked on top of it, the position was valued at about $22.3M to $23.9M, which meant $11.1M to $12M of borrowing power at a 50% collateral factor.
The attacker used it in 18 borrows across the cbBTC, WETH, USDC and wstETH markets, totaling $11,028,762.
595 Liquidations That Came Too Late
Once MAMO's price started falling back, liquidators ran 595 liquidation events and took nearly all of the attacker's mMAMO. That collateral was worth far less than the debt it was meant to cover. The borrowed assets were already gone: Moonwell's post-mortem identified about 8,729,454 USDC burned on Base through CCTP and about 8,728,319 USDC received on Ethereum, where it was converted to DAI and parked in the same wallet that had been funded from Tornado Cash.
Moonwell set borrow caps for every Core Market on Base to 1 wei, which stopped new borrowing across the deployment, and cut the MAMO and WELL supply caps to 1 wei at 11:09:43 UTC. In September it proposed rate changes across seven Base markets that its risk adviser estimated would cut the monthly interest accruing on the bad debt by about 85%, CryptoSlate reported, with protocol reserves earmarked to recapitalize the USDC market.
A Cap on One Door of a Two-Door Room
Every piece of this market did what it was written to do. The supply cap check rejected oversized deposits. The exchange rate math divided the MAMO the contract held by the mMAMO outstanding. The oracle returned the price its source reported. Reviewed one function at a time, none of it is broken.
The danger sits in how the pieces meet. The cap protects the amount of MAMO behind the market, yet it was attached to one of two ways that amount can grow, and the exchange rate reads the total regardless of how the tokens arrived. Add a price source thin enough for the borrower to move, and the market's sense of what its collateral is worth becomes something the borrower writes.
That kind of gap is exactly what scoped reviews are least set up to catch. Listing a new collateral asset usually lands as a parameter change, a cap here and a collateral factor there, on top of market code that was audited on its own. An audit is necessary, and it tells you what a reviewer found inside the scope they were given in the time they had. Nobody was asked to test the new asset's cap, the contract's raw balance and a thin price feed together, and that's the combination that broke.
How Olympix Would Have Prevented It
Olympix reads a market the way an attacker does, as states, transitions and entry points, and it takes in the context around the code, including how assets are listed and capped, to infer the rules the system depends on, and in Moonwell's case they're short. The MAMO backing each mMAMO should only rise through paths the market accounts for, like supply and accrued interest, and never because someone sent tokens to the contract. The collateral a market counts should never exceed the cap it advertises. And no account should be able to borrow more than its collateral can actually be sold for.
The engine would have mapped every entry point that moves the exchange rate: supply, redeem, borrow, repay, and a plain ERC-20 transfer into the contract, which moves the rate without passing through any of the market's own functions. The signal is a cap enforced on the mint path while the value it protects is read from a raw token balance anyone can add to, which is exactly the opening a donation attack uses, as in the $2M Venus Protocol donation attack in March.
BugPoCer would have taken it from there with Counterexample-Guided Invariant Refutation, executing paths from the system outward until one of those rules breaks. The counterexample would have been the August 27 sequence in its smallest form: a small supply, a direct transfer to the market contract, an exchange rate that jumps with no mint, and a borrow that the supplied tokens could never have supported. It would have come back with the conditions that trigger it and a test that replays it.
Had Moonwell run Olympix at every merge and before release, that counterexample would have shown up when the MAMO market was configured, while the fix was still a matter of tracking deposits internally or applying the cap to the balance the exchange rate actually reads. Adding the thin price feed to the same run would have made the economics of the attack visible too, before a borrower got to test them.
Every Way In Counts
Moonwell's cap was a real control on the only door it was watching, and $8.7M left through the other one. For any lending market, a limit that protects a balance has to hold however that balance grows, and a collateral price has to be one the borrower can't move in the same morning they borrow. To see Olympix test your own markets for collateral that can be inflated around its caps, book a technical walkthrough.
On-Chain References
Protocol: Moonwell (MAMO Core Market)
Chain: Base
Date: August 27, 2026, 06:09:45 to 09:30:13 UTC
Loss: about $8.7M bridged out (8,729,454 USDC burned via CCTP); about $9.13M in remaining borrower obligations
Attacker: 0x719e...919D
Funding and receiving wallet: 0xD71d...C384
Direct transfer 1 (34,910,396.68 MAMO): 0x0565...9ce9
Direct transfer 2 (18,482,893.63 MAMO): 0xaf28...3cd9
Largest cbBTC borrow (14.34 cbBTC): 0xafb6...798f
CCTP bridge transfer: 0x9b36...f86e
More post-mortems
See every post-mortem →See what Olympix finds on your own code
Send us a repo that's already been audited, and we'll run Olympix at the same commit and show you what comes back.




