A rental reward function on BNB Smart Chain could pull LULA straight out of the PancakeSwap pair and re-sync it, letting an attacker set the pool's ratio and sell into it. DeFiHackLabs records a loss of about 578.3K USDT.
Security ·
Logged as “2026-07-28: Rental Reward Claims Call Lularecycle Which Moves Up To A Third Of The Pairs Lula To Rental And Syncs With No Usdt Leaving So After A Buy Routed To The Fee Excluded Owner Bypasses Buyenabled The Claims Drain The Pairs Lula Reserve And 4999 Lula Buys Back The Pushed Usdt Plus The Pools Liquidity on LULA”.
LULA's rental programme pays team and referral rewards, and a rental contract sits beside the token to fund those payouts. The function that does the funding is where the incident lives. LULA.recycle() can only be called by that rental contract. It computes a cap of one third of the pair's LULA balance, moves that amount from the PancakeSwap V2 pair to the rental contract using the token's internal _basicTransfer, and then calls pair.sync(). Nothing is swapped and nothing is paid back, so the pair's LULA reserve falls while its USDT reserve never moves. The rental contract is unverified, and DeFiHackLabs infers from the trace that it calls recycle() in a loop of up to ten rounds to cover claims, which means every reward claim takes LULA out of the pool at no cost to anyone.
DeFiHackLabs records a loss of about 578.3K USDT, and its proof of concept replays the attack transaction, mined at block 112,655,390, on a BNB Smart Chain fork with amounts taken from the trace:
_transfer skips that check whenever the sender or the recipient is fee-excluded, so routing the swap output to the owner took the exempt path. The pair goes from 591,841 USDT and 8,001,957 LULA to 197,642,262 USDT and 24,022 LULA.recycle() and then sync(), so the pair's LULA reserve falls from 24,022 to 0.0044 while its USDT reserve still reads 197.64 million.The positions that step 3 claims were not improvised. The proof of concept records that the helper contracts had been accruing team and referral rewards for about twelve days before the attack, and CryptoTimes reported the same finding from CertiK. According to CryptoTimes, TenArmor raised the alert first, with BlockSec Phalcon and CertiK publishing breakdowns on 29 July.
On the figures, DeFiHackLabs puts the attacker's net profit at 578,295.907588061000526778 USDT, reproduced to the wei, and CryptoTimes reported roughly $578,000 with TenArmor at about $578.1K. The borrow is where the sources part company: CryptoTimes put the flash loan at around $237 million, while the DeFiHackLabs proof of concept totals roughly 197.05 million USDT across its fifteen sources, about 40 million less. The token address differs too. CryptoTimes gives one beginning 0x72ad494f, but the chain settles it: 0xF5d7029eb6751d170dcF0Bb1c87Af6f93d5A2e9a, the address in the proof of concept, returns the symbol LULA, is the LULA side of the attacked PancakeSwap pair, and carries the LULA transfers in the attack transaction.
The mechanism is two flaws rather than one. recycle() takes tokens out of an AMM pair without a swap and re-syncs the reserves around the new number, and the token's transfer logic exempts an address from the buy restriction. Either alone is awkward. Together they let an attacker choose the ratio inside the pool and then sell a small amount of token into something the pool believes is almost entirely USDT.
That shape has two precedents in this corpus. CashCowCoin is the closest: a burn function that only the project's trading proxy could call removed CCC straight from the PancakeSwap pair and re-synced it, and the proxy's public sell() called it on anyone's behalf, much as a LULA reward claim makes the rental contract call recycle(). FHToken produced the same reserve mismatch through its sell path, moving FH out of the pair's own balance and calling pair.sync() early. FHToken also gated buys, and its attacker routed the buy output through a whitelisted router, as LULA's attacker routed it to the fee-excluded owner. All three drains ran on flash loans inside a single transaction.
Query every incident yourself — free, no signup
Updated 21 Nov 2026. Figures are as reported by the sources listed above and are not independently verified.