Notional V1
Why it happened
Attack class: Integer Overflow/UnderflowAn unchecked conversion to a smaller integer made a large debt read as zero in the collateral check of the V1 Escrow.
What would have stopped it
- Checked integer conversions
- Retire or drain old deployments
What the score saw
Our pre-hack assessment of Notional treated the first version as retired history and did not check that its Escrow still held funds. It did flag thin development and monitoring activity, which fits an old contract that nobody was watching closely.
Score dimensions
- Maturitymissed
- Operational securitypartly flagged
An attacker took about $1.7 million from the old V1 Escrow of Notional on Ethereum, made up of 69,257.37 DAI and 1,658,524.86 USDC. A number conversion in the collateral check cut a large debt down to zero, so the contract released funds without counting the debt. The Notional multisig paused and upgraded the contract about four hours later. No funds have been recovered, and current Notional versions were not the target.
The attacker prepared the position in tx and drained the V1 Escrow in tx. The multisig paused and upgraded it in tx.
Attack flow
All 4 on-chain stepsThe attack in 2 steps. Step 1, Root control: Notional V1 Escrow (source). Step 2, Drain: Attacker (exit), about $1.73M reached this step.
Step 1: Root control
- SourceNotional V1 Escrow69,257.37 DAI1,658,524.86 USDC0x9abd0b88...f7f683
Step 2: Drain
- ExitAttackerwithdraws against uncounted debt~ $1.73M
Why each step worked
- Unchecked conversion to a smaller integerlarge debt cut down to zero in the collateral check
- Dormant contract still fundedold version held user balances
- No monitoring on the legacy contractmissing controlpause and upgrade came about four hours later
Full forensic detail
Step-by-step reconstruction, root cause, counterfactuals, remediation, and disclosure timeline.
Exploit anatomy
Root cause
Per Notional's post-mortem, the drained contract was the V1 Escrow, an old contract that still held funds. Its collateral check converted a debt amount to a smaller integer type without checking that the value fit. A large enough debt was cut down and read as zero, so the position looked fully collateralised and the Escrow released funds against debt that was never counted. The code was audited and had been dormant, but dormant did not mean empty, and no new review or monitoring covered it. DeFiLlama labels the incident V2, but the post-mortem identifies the V1 Escrow.
Prevention analysis
A conversion that reverts when the value does not fit would have rejected the oversized debt instead of reading it as zero.
Migrating remaining balances out of V1 and closing the Escrow would have left nothing to take.
Alerts on unusual positions or withdrawals from old contracts would have shortened the response from hours.
Similar incidents
Another legacy, audited deployment from an earlier product generation that still held user funds and was drained in 2026 by a numeric edge case in old code.
Remediation
Timeline
Get your protocol scored across 12 dimensions, or request ongoing coverage.