A manipulated BUSD accounting index let attackers borrow 55, 205, 051 KAVA from tokens deposited in Kava Lend’s pool. Bitquery’s transaction analysis says the wallets ultimately kept 53.9 million KAVA. The exploit did not create new coins.
- 127 borrowing rounds took place over 98 minutes on November 14, 2026.
- The supply-index calculation was pushed toward a denominator of just two BUSD base units.
- Bitquery valued the KAVA ultimately kept at about $3.7 million. Other reporting put the loss at about $3.8 million.
- Bitquery reported that KAVA fell 36% in 35 minutes and by half in two hours.
Kava Lend is a lending protocol on the Kava blockchain. The exploit targeted its BUSD market, which Bitquery said was nearly inactive by the time of the attack. Paxos had stopped minting BUSD in February 2023, but the market’s accounting still mattered. Its cash, loans and reserves fed into a live calculation tracking depositors’ claims.
How the accounting flaw worked
Bitquery says the vulnerable calculation used the pool’s cash plus outstanding loans, minus reserves. That figure served as the denominator in a supply-index calculation, which tracks how depositors’ recorded claims change as interest accrues.
At about 04:00 UTC, the pool showed 0.003 BUSD in cash, 5, 084 BUSD out on loan and 7, 618 BUSD in reserves. It also showed 440, 145 BUSD owed to lenders. Based on the cash, loans and reserves, the denominator was negative: 0.003 plus 5, 084 minus 7, 618.
According to Bitquery, the code guarded against a denominator of exactly zero, but not one close to zero. The main wallet first repaid the pool’s 37 open BUSD loans. Then a series of BUSD deposits, loans and withdrawals brought the denominator down to two base units, the smallest BUSD units countable on the chain. Bitquery’s available account of the transactions does not spell out every step or amount in that sequence.
With the denominator near zero, even a small amount of interest could send the supply index soaring. Bitquery says the index rose about 8, 380 times after the initial manipulation, then reached roughly 243, 000 times its pre-manipulation level about three minutes later. The main wallet’s deposit of a few thousand BUSD was recorded as 1.26 billion BUSD in the pool’s accounting.
That was an accounting distortion, not a 243, 000-fold increase in BUSD’s real value or market price. The inflated records made the deposit appear to support far more borrowing than the pool’s actual assets could cover.
Borrowing, liquidations and lender losses
Bitquery traced the exploit to two wallets. One manipulated the BUSD market, while a helper wallet used the distorted accounting to borrow KAVA. The main wallet moved real BUSD out of the pool, leaving little meaningful collateral behind.
Liquidations did not break the cycle. In most rounds, the helper wallet’s loan was liquidated in the next block. But the pool had almost no BUSD to seize. Bitquery says the app put only zero, one or two base units, tiny fractions of a BUSD, up for auction, paid a reward and closed the loan. Closing it freed room under the KAVA loan cap, letting the helper wallet borrow again. That kept the attackers’ sequence going. It did not recover the missing KAVA.
Bitquery counted 127 rounds over 98 minutes, from 04:04 to 05:42 UTC. The helper wallet borrowed 55, 205, 051 KAVA. After repayments and other movements, the two wallets kept 53.9 million KAVA, which Bitquery valued at about $3.7 million at the time.
Other reporting put the loss at about $3.8 million. Available reports do not explain the difference from Bitquery’s $3.7 million estimate. The figures may reflect different quantities or valuation times, so they should not be treated as identical measures.
Bitquery says the KAVA pool fell from 55.3 million KAVA to 519 KAVA during the exploit. The tokens came from assets lenders had deposited in the pool, not from a new issuance. The Kava team reportedly described the incident as “not an infinite mint.”
Price impact and what remains unclear
Bitquery reported that KAVA fell 36% in 35 minutes and by half over two hours on November 14. The Block’s headline described a nearly 40% decline, using a different or unspecified time window. Bitquery said the price held for about six hours after the pool was emptied, then moved as the coins reached BNB Chain.
Bitquery’s tracing identifies a third wallet involved in selling and describes transfers to exchange-related deposit addresses. It also links some later payments to wallets it labels as belonging to HitBTC and Bybit. Those labels and transaction patterns are investigative analysis. They do not establish who controlled the wallets or how much the attackers ultimately realized.
The reported claim that the attackers sold tokens for about $2.1 million is not confirmed by the available transaction analysis. A transfer across a bridge or to an exchange-related address does not, on its own, prove how much was sold or how much the attackers received.
Bitquery’s account also says a bridge blocked 29% of the proceeds it tracked, which had not moved at the time of its investigation. The report does not provide enough detail to reconcile that figure with the $2.1 million sale estimate.
Crypto Briefing reported that Upbit, Bithumb and Coinone froze KAVA transfers that day, and that the Kava team paused a bridge to Crypto Briefing. Available reports do not establish whether Kava Lend’s BUSD market was paused, whether funds were recovered or what fixes the protocol implemented.
The broader lesson is not that an inactive market is harmless, or that every low-volume pool is vulnerable to this exact attack. A quiet market can still carry live accounting assumptions. Reserves, outstanding loans and depositor claims can keep shaping protocol calculations long after ordinary market activity fades. Those calculations need to hold up under adversarial conditions, not just routine use.
Key questions and answers
-
How much KAVA was borrowed and retained?
The helper wallet borrowed 55, 205, 051 KAVA across 127 rounds. Bitquery says the two wallets ultimately kept 53.9 million KAVA after repayments and other movements.
-
Was new KAVA created?
No. Bitquery’s analysis says the tokens came from the KAVA already held in Kava Lend’s lending pool.
-
Why did liquidation fail to recover the loans?
The pool had almost no BUSD to seize. The app closed loans after auctioning just zero, one or two base units of BUSD, freeing room under the KAVA loan cap for more borrowing.
-
How much did the attackers make from selling?
The reported $2.1 million figure is not confirmed by the available transaction analysis, which does not establish the total sold or the attackers’ realized proceeds.