Cascade
Why it happened
Attack class: Oracle manipulationThe attackers held the other side of the vault's shorts, pushed the mark price of thin markets up, and the vault was liquidated into their longs through auto-deleveraging.
What would have stopped it
- Anchor the mark price to an external index, not the thin internal book.
- Cap open interest per account and per market relative to real liquidity.
Not scored before the exploit
BlackHart had not scored Cascade before this exploit.
Attackers on Cascade, a perpetuals exchange on Arbitrum, built long positions in thin markets where the platform's own CLS liquidity vault held the short side, then pushed the mark price up until the vault's positions were liquidated against them. About 1.34 million USDC of vault money was booked to the attackers and withdrawn the same evening, then bridged through Solana and back to Ethereum as DAI. The loss fell on CLS depositors; nobody's wallet keys or approvals were involved. The attackers returned the funds ten days later, Cascade made CLS depositors 100 percent whole, and in September the company announced it was shutting down.
After the mark price was pushed up and the vault's positions were liquidated, the attacker pulled 1.34 million USDC from Cascade's withdrawal contract in one transaction to its wallet, moved it through a relay wallet into the Relay bridge, and later returned the funds to Cascade's recovery multisig.
Attack flow
All 6 on-chain stepsThe attack in 3 steps. Step 1, Root control: CLS vault positions (source). Step 2, Liquidate + withdraw: Cascade withdrawal contract (hub), Attacker wallet (attacker), about $1.34M reached this step. Step 3, Bridge + return: Relay bridge depository (bridge), Cascade recovery multisig (exit).
Step 1: Root control
- SourceCLS vault positionsmain maker in thin marketsshort where the attackers were long
Step 2: Liquidate + withdraw
- HubCascade withdrawal contractvault shorts liquidated, longs deleveraged1.34M USDC paid at 22:34 UTC0x6b01ca28...be99ec
- AttackerAttacker walletreceives the USDCabout 90 minutes after depositing0x285996ba...0d9b55
Step 3: Bridge + return
- BridgeRelay bridge depositoryto Solana, back to Ethereumswapped into DAI0x4cd00e38...38bc31
- ExitCascade recovery multisigfunds returned July 25depositors made whole0x8074583b...152304
Why each step worked
- Mark price set by a thin internal bookthe attackers could move the price that judged the vault
- No open interest cap against the vaultmissing controltwo accounts built positions large enough to matter
- No hold on fresh deleveraging profitmissing controlwithdrawal cleared at once for a brand-new account
- Invite chain and exchange trailSEAL 911 traced activity to Polymarket and Binance
Full forensic detail
Step-by-step reconstruction, root cause, counterfactuals, remediation, and disclosure timeline.
Exploit anatomy
handleOps((address,uint256,bytes,bytes,bytes32,uint256,bytes32,bytes,bytes)[],address)tx:0xcd801d24...578314Root cause
The CLS vault was both the main liquidity provider in Cascade's thin markets and a leveraged counterparty in them, and the price used to judge its solvency was a mark price that could be moved inside those same thin markets. That combination let a trader who held the other side of the vault's positions also move the price that decided whether the vault was liquidated. Per Cascade's disclosure, the attackers accumulated longs where CLS was short, pushed the mark price up, and the vault's shorts were liquidated or closed out while the attackers' longs were auto-deleveraged at the inflated mark, turning the price move into about $1.34 million of realized profit that was withdrawn about 90 minutes after the attackers deposited. No contract was hacked, no key was stolen and no flash loan was used; the failure was market design. There was no cap on how much open interest one account could build against the vault in an illiquid market, no reference index outside the thin order book to anchor the mark price, and no hold on large, fresh profits from liquidation and deleveraging events before they could be withdrawn. Cascade said it was still investigating the exact mark price mechanism. Analysts had warned depositors in the weeks before, and vault liquidity reportedly fell by more than 95 percent before the attack, which left the remaining markets even thinner.
Prevention analysis
A mark price built from a median of outside venues, with a cap on how far it can stray from that index, cannot be pushed up by trading in a market where the vault is almost the only maker. The vault's shorts would not have been liquidated at the manipulated level.
The attack needed one or two accounts to hold positions large enough to matter against the vault. Position limits tied to market depth would have kept the attackers' longs, and the profit available from deleveraging them, far smaller.
The $1.34 million left about 90 minutes after the attackers deposited. A short delay on withdrawals of newly realized profit from new accounts would have let the team pause before the money was bridged out.
Similar incidents
A trader moved the price of a thin market they were positioned in, and the protocol valued their position at the manipulated price and let them withdraw against it. Same root cause: a manipulable mark price used to settle real value.
A community liquidity vault inherited a large short in an illiquid perpetual market after an attacker squeezed its price, the same vault-as-counterparty exposure that cost the CLS vault here.
Remediation
Timeline
Get your protocol scored across 12 dimensions, or request ongoing coverage.