A PancakeSwap pair's public skim() and sync() let an attacker collapse the NEX side of the NEX/AIC pool with a flash swap, then buy out nearly the whole AIC reserve at the wrecked ratio. DeFiHackLabs reports a net gain of about 32.36 BNB, roughly $21.5K, on BNB Chain on 3 August 2026, a date confirmed from the transaction's own on-chain record.
Security ·
Logged as “2026-08-03: Pair Skim Reserve Mismatch Exploit Flash Swap Leveraged on AIC”.
AIC and NEX are both traded on PancakeSwap V2 pairs on BNB Chain, NEX only against AIC and AIC against both NEX and USDC, and a Uniswap-V2-style pair's skim() is a public function that pays any caller the difference between a token's real balance and the pair's recorded reserve. That public access is what the attacker at 0xC3cB0872C42BFA5EB3B0258D7EEA2cCaF6a49475 (shortened 0xC3cB…9475) used to empty the NEX/AIC pair at 0x974C0078740480aE830D379fDB8d5f441C9dDC75 on 3 August 2026. The attack transaction, 0x905cc861bcc525d3a8e699583943831b97500bbac11c92dc20ed6edbddd69f87, was sent from that address, mined in block 113782392, and timestamped 13:03:53 UTC on 3 August 2026, matching the date in DeFiHackLabs's README heading. The transaction carries no to address, so it is a contract creation. (The published proof of concept deploys its own AICAttack contract and then calls attack(); it does not reproduce a constructor-based attack.)
The sequence, reconstructed by DeFiHackLabs from the on-chain trace:
0xe89636FB73D04Db51e5Fbd0Ce1379fb8d2b96415, minus one wei.pancakeCall callback, swap all the borrowed AIC into NEX through the NEX/AIC pair, piling AIC into that pool.skim(), paying the surplus NEX to the router, then sync(), which locks the pair's recorded reserves at roughly 1 NEX against roughly 83.26 million AIC.DeFiHackLabs reports a net gain of about 32.36 BNB, roughly $21.5K, from that transaction. A residual balance of roughly 2 million NEX ended up at a separate helper address, 0x0f7e35653f6A8E09A0865a183B51177e16237CB5, which the reconstruction treats as outside the attacker EOA's BNB gain.
The proof of concept's own comments describe both AIC and NEX as fee-on-transfer, but that does not hold for AIC under its own code. AIC's contract was verified on Sourcify on 17 August 2026, two weeks after the attack, as a flattened OpenZeppelin ERC-20 with a 210,000,000-token mint at deployment and a claimStuckTokens recovery function. It carries no transfer tax or fee logic of any kind. NEX, verified on Sourcify the same day, is the one that taxes transfers: its `_transfer` override deducts a daoFee and a nodeFee, 3 percent each as deployed, whenever a transfer lands on an address flagged in automatedMarketMakerPairs, a flag its constructor sets for the pair it creates against AIC, the same NEX/AIC pair the attacker later drained. Any NEX moving into that pair, by a router swap or the plain transfer in step three above, is taxed before the pair's balance reflects it. The pair contract itself, the one the attacker actually calls skim() and sync() on, remains unverified, so the missing access check on those two functions is established from the transaction trace rather than from the pair's own source.
DeFiLlama's hacks database has no entry for this incident, and no security firm, news outlet, or either project has published an analysis of it. Beyond the two token contracts' verified source and the attack transaction's own on-chain record, the loss figure and the exact call sequence rest on DeFiHackLabs's reconstruction alone.
A comparable bug surfaced three weeks later on a different token. FHToken's sell-tax reserve mismatch via premature sync is the same family of fault, a pair's recorded reserve forced out of step with its real balance through a permissionless call, then sold against at the resulting price. Neither case needed a bug in skim() or sync() themselves. Both needed only that the functions were public.
What would have stopped it is simple in both cases: restrict skim() to the pair's own LP-token holder, or build in a floor that keeps the recorded reserve close to the live balance regardless of who calls it. A pair that trusts its own balance only after a swap, mint or burn, never after an arbitrary transfer followed by a bare sync(), removes the mismatch these attacks depend on. NEX's own transfer tax, which exists to fund its dao and node addresses rather than to protect against this, offered no defense here: the pair-level functions it moved tokens through carry no such check at all.
Query every incident yourself — free, no signup
Updated 8 Nov 2026. Figures are as reported by the sources listed above and are not independently verified.