On 18 November 2026 at 11:41:38 UTC, the USDT0 aToken behind Reental’s RNT Lend market on Polygon emitted an Upgraded event. Five seconds later, the USDC aToken was upgraded to the same implementation, 0xcc555fac…ac09. Twenty seconds after the first upgrade, 294,040.356612 USDT0 left the USDT0 aToken. Four seconds after that, 160,785.04943 USDC left the USDC aToken. Those two withdrawals total 454,825.406042 stablecoins. Blockaid disclosed an ongoing exploit at 12:39:42 UTC, said an admin had upgraded the USDT and USDC aToken proxies “to a contract that lets anyone drain the underlying tokens,” and reported about $454,000 drained across two transactions. Onchain Lens independently reported about $454,000. A later review by Bitquery, also recorded by DefiLlama, raised the stablecoin total to $954,703: the aToken cash plus about $499,877 taken from user wallets and Safes. Those figures describe different slices of the same morning and must not be added together a second time. This is a defensive account of a public incident, not a reconstruction of how to repeat it, and it does not establish that remaining collateral or other Reental products are safe.
RNT Lend is Reental’s lending market on Polygon. Reental, a Spanish firm that sells tokenized real estate, uses the market so lenders can deposit dollar stablecoins and borrowers can lock supported property tokens as collateral. DefiLlama describes it as an Aave V3 fork. Reental’s own account has said RNT Lend “uses Aave V3-based technology on Polygon.” Before the drain, Bitquery counted 105 property markets and two dollar markets: one for USDT0, Tether’s dollar token on Polygon (Blockaid referred to it as USDT), and one for USDC. Each market holds its cash in an aToken — a receipt token that also holds the underlying coins.
Blockaid’s alert and follow-up listed:
0xB0fE6E9c4EA2d567E1fb6e45ecF2c9ddB34011010xb977d09c…642f0xfaF2338E041258A43c062d60296A9F8f0F8C21210xd3cb693d157816dd4a9de27b62c6e723db49d880Bitquery later identified the address Blockaid labeled the exploiter as the sender of an MEV bot that took the USDT0 cash, not as the admin keyholder. The two are different actors.
The first Upgraded event is on the USDT0 aToken in transaction 0xa97dcfe2…7e93, in block 97531086, at 11:41:38 UTC. The second is on the USDC aToken in transaction 0xdd760fef…e84d, in block 97531089, at 11:41:43 UTC. Both calls were sent from 0x30729b69…2691 to 0xb67ed59e…3d45 and pointed both proxies at 0xcc555fac…ac09.
The USDT0 withdrawal is Blockaid’s example transaction, in block 97531099, at 11:41:58 UTC. The sender was 0xB0fE6E9c…1101. Official Polygon USDT0 at 0xc2132d05…8e8f moved 294,040.356612 USDT0 from the aToken to 0x30178988…44b9. We called the aToken’s USDT0 balance historically on an archive node: at block 97531098, the last block before that transaction, it held exactly 294,040.356612 USDT0. After the call, that cash was gone.
The USDC withdrawal is transaction 0x60674299…fa79, in block 97531102, at 11:42:02 UTC. The sender was 0xde0a99fb…1163, the address Bitquery labels as the USDC MEV bot. Native USDC at 0x3c499c54…3359 moved 160,785.04943 USDC from the USDC aToken to 0x3cb2452c…e520. The aToken’s USDC balance at block 97531101, the last block before that transaction, was exactly 160,785.04943 USDC.
The two withdrawals sum to 454,825.406042 stablecoins, matching Bitquery’s $454,825 and sitting in the same range as Blockaid’s and Onchain Lens’s $454,000 snapshots. When we checked later, both aTokens held 0 of their underlying stablecoins.
Bitquery wrote that the new implementation paid the market’s coins to any address the caller named, and that the keyholder’s own calls in those same blocks failed after the bots went first. We confirmed that the cash left through the two successful bot transactions, and that Bitquery’s bot sender addresses match the senders we observed.
Blockaid’s first figure covered the aToken cash. It was not the full stablecoin outflow.
We followed USDT0 into the collecting wallet Bitquery attributes to the keyholder, 0xda45c0b3…6c19. Between the aToken drains and block 97531778, that wallet received 499,876.377308 USDT0 across 1,164 transfers. A second wallet, 0x9ff1a6a5…83b8, then received 499,877.376992 USDT0, almost all of it from the collecting wallet, and sent that exact amount onward: 10,100 USDT0 to Bridgers at 0xc1d13492…a1b5 and 489,777.376992 USDT0 to Relay at 0x4cd00e38…bc31. Bitquery reported $499,877 taken from 1,149 user wallets made by Reental’s wallet factory and from four Safes, then bridged off Polygon within the hour. Our observed flows match that figure. Adding them to the aToken drains produces about $954,702, in the same range as Bitquery’s and DefiLlama’s $954,703.
Bitquery also reported that the keyholder moved about $49.34 million in property tokens, valued at the pool’s own prices, and that Reental pulled every one of those tokens back within about two hours through its token manager. DefiLlama records the same recovery. We did not independently re-price those tokens. That movement is separate from the $954,703 stablecoin total.
According to Bitquery, lenders had $10.01 million in the two dollar markets before the drain, borrowers had $9.57 million of it on loan, and less than 5% sat as cash — the $454,825 the bots took. Outstanding loans do not disappear when the cash buffer is emptied. Whether, and how much, lenders can withdraw is a question only Reental can answer.
Bitquery’s 19 November update, checked at 12:47 UTC, said Reental removed the implicated key from the Safe that owns the pool at 14:03 UTC on 18 November, paused all 108 markets at 16:55 UTC, and replaced the code behind all 91 markets changed during the drain at 23:05 UTC. Deposits, repayments and withdrawals still would not go through at that check. We found no public incident statement from @Reental_co or @rntprotocol as of writing; the official Reental account continued product-marketing replies about RNT Lend on 19 November.
DefiLlama classifies the incident as a compromised private key. The chain shows that a privileged admin path upgraded the aTokens and that a key Bitquery traces to Reental’s own history signed the morning’s setup steps. As Bitquery notes, a blockchain cannot tell a stolen key from an insider: both sign the same way. Reental has not said which it was.
Aave V3 itself is not the failed component in this account. The failure sits in how this market’s upgrade authority was held and used — the same class of risk as when Yapeswap’s factory changed hands through a privileged path, and adjacent to the leftover approvals behind the Pennysia settlement drain and a stale Safe approval on Avalanche.
The sketch below is illustrative, not RNT Lend or Aave source, and is not a reproduction of the incident. It only states a property an aToken withdrawal path should not be able to skip:
// Simplified, illustrative invariant — not RNT Lend or Aave source.
// An upgraded aToken must not pay underlying reserves to an arbitrary caller.
function withdrawUnderlying(address to, uint256 amount) external {
require(msg.sender == address(pool), "not authorized");
IERC20(underlying).transfer(to, amount);
}
Treat the upgrade key as a production credential, not a convenience. If a single key can replace the code behind a market that holds deposits, that key is a freeze-and-drain switch. Hold it on a high-threshold Safe, put a timelock in front of implementation changes, and monitor Upgraded events so a surprise implementation is visible in seconds.
Do not assume an audit of the current implementation covers tomorrow’s implementation. Forking Aave V3 can give a market a well-reviewed base, but it does not constrain what an admin key may point the proxies at later. An upgrade that removes access control is a privilege being used, not a novel bug class.
Separate emergency pause from upgrade-and-replace. A pause that takes hours, after cash has already left and after a permissionless implementation has been live, is a late control. The pause, the key removal and the later code replacement reported by Bitquery mattered for containment. They did not prevent the $954,703 outflow.
Remember that a thin cash buffer can vanish in one block. A lending market that pays a double-digit supply rate will lend out most deposits. Once the residual cash is gone, remaining “TVL” can still look large because loans and collateral remain on the books.
If you deposited USDT0 or USDC into RNT Lend, or used Reental property tokens as collateral, follow updates only from the project’s own channels, beginning with Reental on X and RNT Protocol. Treat unsolicited “refund” or “recovery” messages as phishing until the project publishes a process.
The property tokens Bitquery and DefiLlama say were moved and later returned are not a promise that stablecoin deposits can be withdrawn. Do not deposit additional funds until the team has published a post-mortem and stated that the markets are safe to use. Revoking approvals you granted to RNT Lend contracts cannot undo a transfer that has already left, but it can limit later misuse of an allowance.
0xB0fE6E9c…1101, example transaction 0xb977d09c…642f, victim aTokens 0xfaF2338E…2121 and 0xd3cb693d…d8800x30178988…44b90x3cb2452c…e520Upgraded to 0xcc555fac…ac09Upgraded to the same implementationThe RNT Lend loss did not require a failure of Polygon or of Aave V3’s core contracts. An admin upgrade replaced the code behind two dollar aTokens, 454,825.406042 stablecoins left those contracts in two transactions, and about $499,877 more USDT0 followed from wallets and Safes before the key was removed and the pool was paused. If you are building a lending market whose reserves sit behind upgradeable proxies, talk to our team or check our audit pricing to have upgrade authority, timelocks and aToken access control reviewed before a single admin transaction is the first production test of those privileges.
K3 Capital paused sBOLD after the Ethereum vault's totalAssets fell by 3,428,799 BOLD in minutes. Blockaid reported about $3.43 million drained.
A single Base transaction sent 404,540 USDC out of B18TimeBank's settlement contract after signed level-reward vouchers were redeemed.
A MORE Markets liquidation bug on Flow EVM cleared collateral without the matching WFLOW debt, emptying the reserve. Blockaid estimated about $308k.
Independent reports say interest-index math in Kava Lend's idle BUSD market inflated collateral, after which about 55 million KAVA was borrowed. Kava has not published an official loss figure or incident report.