On 14 November 2026, starting around 04:01 UTC, Kava Lend’s nearly idle BUSD market became the entry point for a drain of the protocol’s KAVA deposits. Independent on-chain reviews by The Block and Bitquery say roughly 55.2 million KAVA was borrowed against inflated BUSD collateral. At a pre-selloff price near $0.068, that is about $3.7–$3.76 million. DefiLlama records the stolen amount as about 53.9 million KAVA (~$3.7 million), which matches Bitquery’s figure for what the attackers kept after other loans in the market were repaid.
Kava has not published an official loss figure or a post-mortem. As of 16 November 2026, 01:15 UTC, the project’s official account had not followed its 14 November promise of a complete incident report, and we have not seen an announced code fix. The title figure of 55 million KAVA is therefore a reported total, not an official one. This article stays with those public reports, Kava’s own statements, and the protocol’s published interest-index code. It is a defensive account of the accounting failure, not a reconstruction of how to repeat it.
Kava Lend, also documented as the HARD module, is the chain’s money-market: suppliers deposit assets, borrowers post collateral, and each market tracks accrued interest with a supply interest index. When that index rises, every supplier’s recorded deposit of that asset is marked up by the same factor.
The market involved was BUSD, the Binance-branded dollar stablecoin that Paxos stopped minting in 2023. By mid-November the market was effectively dormant. Independent reviews describe a book that no longer looked healthy: a large amount still owed to BUSD suppliers, little cash in the pool, and reserves that were large relative to cash and outstanding borrows. That kind of stale market is dangerous because the interest index is updated by dividing newly accrued interest by net supply:
net supply = cash + borrows − reserves
If that denominator is robust, each accrual adds a small, bounded increment to the index. If it collapses toward zero, the same slice of interest can reprice every BUSD deposit by an enormous factor. The public source in x/hard/keeper/interest.go already skips the division when net supply is exactly zero and returns a factor of one. It does not treat a tiny positive net supply the same way.
Reports agree on the effect, even though Kava has not confirmed the arithmetic itself: over a few minutes on 14 November, the BUSD supply index was described as rising by about 243,000 times. Deposits that the protocol already treated as BUSD collateral were then large enough, on the books, to back repeated KAVA borrows from the main lending pool.
The Kava Lend module account that holds pooled KAVA deposits was reported to have fallen from about 55.33 million KAVA to a residual cash balance during a window of roughly 90–98 minutes. Bitquery recorded about 519 KAVA left immediately after that window, and later the same day measured about 0.19 KAVA. The Block also uses the 0.19 KAVA figure. In either reading, almost all of the KAVA that suppliers were owed was gone.
Kava’s later clarification is consistent with that picture. The team said the event was “not an infinite mint” and was “isolated and contained within Kava Lend.” New KAVA was not created. Existing deposits in the lending module were borrowed out against collateral the protocol believed was there.
Times below are in UTC, compiled from Kava’s posts and from the independent reviews cited in Sources. Where those reviews disagree on a residual balance or a sale breakdown, both figures are kept.
Secondary reporting named Upbit, Bithumb and Coinone as the Korean exchanges that suspended KAVA transfers. Kava itself did not name those venues. Transfer status can change, and users should check the relevant exchange directly.
The failure belongs to a familiar family: an internal index, share or exchange-rate is allowed to disconnect from the assets actually in the pool, and a later action — here, borrowing against the marked-up deposit — spends other users’ funds. We have described the same class of bug in the ACryptoS depositMulti vault exploit and the Arcus Perps subaccount accounting exploit.
The relevant public function is short. The snippet below is the published CalculateSupplyInterestFactor logic, simplified only in naming, and a defensive guard that the current code does not apply. It is not a reproduction of the incident and omits market-state inputs:
// Simplified from x/hard/keeper/interest.go
func supplyFactor(newInterest, cash, borrows, reserves Dec) Dec {
netSupply := cash.Add(borrows).Sub(reserves)
// Existing guard: exact zero only
if netSupply.IsZero() {
return One
}
// Missing in the published code:
// a tiny positive netSupply still goes through the division.
// A defensive version would also reject or cap this case, e.g.:
// if netSupply.IsNegative() || netSupply.LT(minNetSupply) { return One }
// or: factor = min(newInterest/netSupply+1, maxFactorPerAccrual)
return newInterest.Quo(netSupply).Add(One)
}
Two conditions made that omission expensive. First, an idle market can sit for years with reserves, leftover debt and almost no cash, so net supply is already fragile. Second, the index is a global multiplier: once it jumps, every remaining BUSD claim is revalued together, including deposits the protocol will later accept as collateral.
A borrow cap on KAVA is not a substitute for that invariant. Independent reviews say a cap was in place and that KAVA was still borrowed out over many separate loans until the pool’s cash was effectively gone. The cap limited how much could be taken in one step. It did not keep the module’s assets in line with its books.
Kava has not confirmed this reading. Bitquery is explicit that its account comes from public nodes and open code, and that the team’s forthcoming report may describe the cause differently. Until that report appears, the defensive lesson does not depend on any single reconstructed input: interest and exchange-rate math must reject or bound a near-zero denominator, not only an exact zero.
Kava’s public posture on 14 November had three parts: the issue was confined to Kava Lend, it was not an infinite mint, and a full incident report would follow. The team also paused Axelar withdrawals and said partner exchanges had paused deposits and withdrawals. We have not seen an official reimbursement plan, a named attacker address from the project, or an announcement that Kava Lend itself was paused or patched.
On the movement of funds, the independent reviews are directionally consistent and differ in some breakdowns:
BLOCKED_WALLET). DefiLlama also records that blocked amount. Bitquery’s residual on-chain positions and later hops were last checked on 14 November and may have changed.Those figures describe reported proceeds, not a final loss accounting for suppliers. The protocol still owed KAVA to lenders after the cash had left the module. Whether suppliers will be made whole is a question for Kava’s promised report.
A separate caution from Bitquery, as of its 14 November snapshot, is that the BUSD market may have remained in an unsafe state, with an inflated claim still on the books. We have not independently re-checked that state, and we are not testing it. Until Kava publishes a fix and an all-clear, users should not assume that depositing into Kava Lend — especially the BUSD market — is safe.
This article is technical commentary, not financial advice.
Guard near-zero denominators, not only exact zero. if supply == 0 is not enough for an index, exchange rate or utilization ratio. Dust, rounding and a market whose reserves already exceed cash plus borrows can all produce a tiny or negative net supply. The accrual path should refuse to update, freeze the market, or cap the factor when the denominator is below a conservative minimum.
Idle markets are an active risk. A market can stay listed long after its asset is no longer minted or traded. Governance should retire, pause or tightly cap markets whose cash, borrows and reserves no longer support a stable index. An unused market is still in the same collateral system as the assets you care about.
Cap how far an index can move in one accrual. Even with a healthier denominator, a maximum per-block or per-second factor would have turned a catastrophic reprice into a bounded accounting error. Alerting on sudden index jumps is the operational counterpart of that cap.
A recorded claim must still be covered by assets. Once a supply index has been allowed to run ahead of the pool, later borrows, withdrawals and liquidations will all trust the wrong number. Invariant tests should assert that no sequence of accruals can increase a supplier’s claim except by interest actually paid, within a bounded rate, and that liquidation cannot close a debt without recovering comparable collateral.
Borrow caps and per-call limits are throughput controls, not solvency controls. They slow an exit. They do not repair a collateral index that has already detached from the pool.
If you supplied assets to Kava Lend, especially KAVA, follow updates only from the official account KAVA_CHAIN and from Kava’s own site. Treat unsolicited “refund,” “claim” or “compensation” links as phishing until the team publishes a report and a recovery process.
Independent reviews say the KAVA cash left in the module was negligible compared with what lenders are owed, so withdrawals may not be fully available. That is a description of reported pool balances, not a prediction of how Kava will treat suppliers.
If you hold KAVA in a wallet or on an exchange and were not a Kava Lend supplier, the published accounts do not say your tokens were taken from those venues. The market price did fall sharply after the BNB Chain sale. Exchange deposit and withdrawal status is controlled by each venue and may still be restricted.
Do not deposit into Kava Lend until the project has published a fix and confirmed that affected markets are safe to use.
CalculateSupplyInterestFactor divides new interest by cash + borrows − reserves and short-circuits only when that total is exactly zeroThe Kava Lend loss is a reminder that a money market can be emptied without an infinite mint and without a new token being created. If interest-index math can reprice collateral in an idle market, the assets sitting in healthier markets become withdrawable against a number that is no longer real. If you are building or maintaining a lending protocol, talk to our team or check our audit pricing to have denominator guards, idle-market policy and index-growth caps reviewed before a stale market is the first production test.
An attacker altered Pool Party snapshot records, deposited 10 USDC, and drained 15 pools on Arbitrum and Base for about $215k.
How an unchecked venue call in Pennysia's Settlement contract on Ethereum let an attacker spend users' leftover token approvals, draining four wallets.
How an attacker used a flaw in the skew calculations of Rujira's concentrated liquidity ranges to unbalance BTC/USDC positions on THORChain for about $200K.
How a subaccount transfer accounting bug let one new account inflate its balance and withdraw 2.23M USDG from Arcus Perps on Robinhood Chain, and what stopped more.