BlackHart
Go live
Hacks feed/Moonwell

Moonwell

August 27, 2026·Base·Oracle Manipulation
$8.7M
total loss
StatusConfirmed
View the Moonwell score →

Why it happened

Attack class: Oracle manipulation

Direct MAMO transfers inflated the exchange rate while thin-market buying pushed the price feed up, so illiquid collateral backed liquid loans.

What would have stopped it

  • Cap collateral exposure for thin tokens to what the market can absorb
  • Smooth the collateral price with a time-weighted average and a deviation limit

What the score saw

758Damascus, scored May 15, 2026

Our pre-hack assessment of Moonwell rated its price-feed design and economics as middling but did not call out thinly traded collateral or the direct-transfer route into a market's accounting. That is the gap this attack used.

Score dimensions

  • Oracle integritymissed
  • Economic soundnessmissed
  • Adversarial resiliencemissed

On August 27, 2026 an attacker used the thinly traded MAMO token to borrow about $11 million of cbBTC, ETH, USDC and wstETH from Moonwell on Base, leaving about $8.7 million of lasting damage per DeFiLlama and press reports. They sent tens of millions of MAMO straight into the market to inflate their collateral and bought MAMO on a shallow market to push its price up about forty times. Liquidators recovered part of the debt, and Moonwell's post-mortem puts the remaining borrower obligations near $9.1 million at its evidence cutoff, an estimate rather than a confirmed loss. The proceeds were moved to Ethereum as DAI, and Moonwell froze new borrowing the same morning.

Moonwell markets on other chainssafe
Moonwell governance and stakingsafe
Base core markets for cbBTC, WETH, USDC and wstETHdrained(about $11 million borrowed at the oracle price, lasting loss about $8.7 million)
Depositors in the Base USDC core marketdrained(treatment under discussion in governance)
Exploit anatomy

An attacker-linked account funding wallet moved capital to Base and passed MAMO and USDC to a principal account. The principal account used the MAMO collateral market, sent MAMO to it directly in one transfer and another, and borrowed liquid assets against the inflated collateral. The proceeds left Base in two CCTP burns, the first and the second, and were swapped to DAI in this transaction.

The attack in 3 steps. Step 1, Inflated collateral: mMAMO collateral market (source). Step 2, Drain: Principal account (hub), about $11.03M reached this step. Step 3, Exit: Ethereum DAI (exit), about $8.73M reached this step.

  1. Step 1: Inflated collateral

    • Source
      mMAMO collateral market
      53.39M MAMO sent directly
      exchange rate up about 3.7x
  2. Step 2: Drain

  3. Step 3: Exit

Why each step worked

    • Thin MAMO marketprice pushed from $0.0106 to $0.431
    • Direct transfers inflated the ratebypassed the supply cap
    • No collateral cap tied to depthmissing controlilliquid token backed liquid loans
    • No averaged pricemissing controlspot-driven feed accepted
    • Caps lowered after 80 minutesmissing controlno automatic borrow limit
Full forensic detail

Step-by-step reconstruction, root cause, counterfactuals, remediation, and disclosure timeline.

Exploit anatomy

1.
Between August 21 and 23 an account linked to the attacker received about 800 ETH from Tornado Cash, sold it for about 1.95 million USDC, and moved the USDC to Base. It bought about 7.4 million MAMO, then sent MAMO and USDC to a second account, per the Moonwell post-mortem.
2.
From 06:18 UTC on August 27 the second account minted mMAMO, Moonwell's MAMO collateral receipt token, and posted it as collateral. It began borrowing at 08:43 UTC.
3.
At 09:19 and 09:21 UTC the attacker sent about 53.39 million MAMO directly to the mMAMO contract without minting, which raised the market's exchange rate from about 0.0205 to 0.0755 and so the value of the existing collateral. Thin on-chain buying pushed the MAMO price feed from about $0.0106 to $0.431 by 09:28 UTC.
4.
By 09:30 UTC the attacker had borrowed cbBTC, WETH, USDC and wstETH worth about $11.03 million at Moonwell's own price at each borrow block. Liquidators began repaying part of the debt from 09:30 UTC onward, which is why the lasting loss is lower than the gross figure.
5.
At 09:41 and 09:42 UTC about 8.73 million USDC was burned on Base through Circle's CCTP in two equal parts and arrived on Ethereum, where it was swapped into about 8.73 million DAI.
6.
At 10:53 UTC Moonwell lowered the borrow cap of every core Base market to 1 wei, and at 11:09 UTC cut the MAMO supply cap to 1 wei, stopping further borrowing.

Root cause

Moonwell let a thinly traded token, MAMO, back large loans of liquid assets, and the lending market trusted two numbers the attacker could move. Per the Moonwell post-mortem on its governance forum, the incident combined collateral-accounting inflation with price manipulation. Sending MAMO straight to the mMAMO market contract, with no minting, raised the exchange rate that converts the attacker's mMAMO into MAMO, and that route did not consume the market's supply cap. At the same time, buying MAMO on a shallow on-chain market pushed the price feed the lending market reads from about one cent to about 43 cents. Together they inflated the collateral value, and the attacker borrowed cbBTC, WETH, USDC and wstETH against it. No key was stolen and no contract was broken in the usual sense: the market valued collateral using inputs that were cheap to move relative to the loans they released. The attacker spent roughly $2 million of initial capital and about $7.5 million of gross MAMO purchases, much of it recycled borrowed funds, to take $11 million of gross borrows, per the post-mortem.

Prevention analysis

Cap collateral exposure for thin tokens to what the market can absorb

A supply cap tied to real market depth, enforced on every path into the market including direct transfers, would have limited the value that MAMO could back.

Smooth the collateral price with a time-weighted average and a deviation limit

A price that cannot move from one cent to forty-three cents inside a few minutes would have kept borrowing power flat while the pump happened.

Rate-limit new borrowing against newly listed or illiquid collateral

About two and a half hours of steady borrowing passed before caps were lowered. An automatic per-market borrow limit would have halted it much sooner.

Similar incidents

Tectonic (2026)

A thinly traded governance token was pumped and used as collateral to borrow far more liquid assets from the lending markets.

Nostra Money Market (2026)

An illiquid token's price in a shallow pool was fed into the lending market and borrowed against.

Mango Markets (2022)

A low-liquidity asset's price was pushed up and used as collateral to drain the lending pool.

Remediation

1.Moonwell lowered the borrow caps of every core Base market and the MAMO supply cap to 1 wei on the day of the attack to stop further borrowing.
2.Anthias is preserving the evidence, and Moonwell says recovery and remediation updates will follow in the governance forum. Treatment of depositors and the MAMO market is still being discussed in governance, and remediation is pending.pending
3.Other lending markets should cap thin-token collateral to market depth, count direct token transfers in supply accounting, and use averaged prices for long-tail assets.

Timeline

2026-08-27First transaction by the attacker's principal account.
2026-08-27First borrow, 0.5 cbBTC.
2026-08-27First direct MAMO transfer to the mMAMO contract.
2026-08-27MAMO price feed peaks at about $0.431.
2026-08-27Final successful borrow, 990,000 USDC.
2026-08-27First of two CCTP burns moving USDC from Base to Ethereum.
2026-08-27Moonwell sets every core Base market's borrow cap to 1 wei.
2026-08-27MAMO supply cap set to 1 wei.
Moonwell and Anthias (post-mortem). ExVul (price analysis)
Continuous adversarial monitoring

Get your protocol scored across 12 dimensions, or request ongoing coverage.