Skip to content
Oblique view of a dark wall of identical teal gauges; one red gauge's needle has wound past its intact zero stop pin, while a small silhouette looks up at the healthy dials.
exploitsNovember 16, 20263 min read

Kava Lend's $3.76M Drain: Two Base Units of BUSD Supply Inflated Deposits 243,000x

Dmitry Serdyuk
Dmitry SerdyukCo-Founder & CDO

Updated on November 16, 2026

TL;DR

At 04:01 UTC on November 14, 2026 (11:01 p.m. US Eastern on Friday), an attacker pushed the net supply of Kava Lend's dormant BUSD market down to two base units, two hundred-millionths of a dollar. Kava Lend pays depositors by dividing borrower interest by that net supply, and its code only refuses the division when supply is exactly zero. Within three minutes the BUSD supply index rose roughly 243,000-fold, and the attacker's earlier BUSD deposit counted as about 1.25 billion BUSD of collateral, according to The Block's review of onchain data. A second wallet then borrowed KAVA 127 times, and the first wallet liquidated each loan so the debt vanished and the KAVA borrow cap reset. The total was about 55.2 million KAVA, worth roughly $3.76 million at the pre-attack price of $0.068. About 49.2 million KAVA left through LayerZero to BNB Chain and sold for about $2.09 million. Kava says the incident was "not an infinite mint" and is contained to Kava Lend. As of November 16 it has published no loss figure and no post-mortem. We rebuilt the core arithmetic from Kava's public REST API and source code, and the numbers match.


What Happened to Kava Lend?

Kava Lend is the native money market of the Kava Layer 1, a Cosmos SDK chain, and runs as a Go chain module (x/hard). The module account holding its pooled KAVA went from 55.33 million KAVA to 0.19 KAVA in a little over an hour and a half, per The Block. That was the low point.

The attack ran through the market for BUSD, the Binance-branded stablecoin that Paxos stopped minting in 2023. The Block describes the market as all but empty before the attack. By our queries, the part that was nearly empty was its net supply after reserves, while about 449,715 BUSD of deposits stayed on record. Blockchain analytics firm Bitquery traced the operation to two wallets and counted 127 borrowing transactions over about 98 minutes, as reported by FinanceFeeds. We found no direct link to Bitquery's own write-up, so its figures here are cited as relayed.

Kava's first public word came at 6:28 a.m. ET on Saturday, more than seven hours in, The Block reports. About half an hour later the team wrote that "This is not an infinite mint" and that the event was "isolated and contained within Kava Lend." Upbit, Bithumb and Coinone suspended KAVA deposits and withdrawals, and Upbit put the token on its delisting watchlist, per CoinCentral. KAVA fell about 40% on the day.

The Block also reports repeated deposit and borrow cycles in the BUSD market from November 8, six days before the drain.


How Does Kava Lend Pay Its Depositors?

Kava Lend tracks depositor earnings with a supply interest factor, one number per market. Each time interest accrues, the module takes the interest borrowers owe, subtracts the reserve cut, divides the remainder by the market's net supply, and multiplies the factor by one plus that ratio. A depositor's balance grows by the same ratio the factor grew since they last synced. If the factor doubles, every deposit in that market doubles on paper.

Net supply is cash + borrows - reserves: BUSD sitting in the module, plus BUSD lent out, minus the protocol's own reserve claim. The function sits in the x/hard interest keeper in Kava's public repository:

func CalculateSupplyInterestFactor(newInterest, cash, borrows, reserves sdk.Dec) sdk.Dec {
	totalSupply := cash.Add(borrows).Sub(reserves)
	if totalSupply.IsZero() {
		return sdk.OneDec()
	}
	return (newInterest.Quo(totalSupply)).Add(sdk.OneDec())
}

The only guard is for zero.

The same file computes the utilization ratio from the same denominator, and that function does treat a tiny or negative supply as a hazard: it returns 100% when supply goes negative and caps the ratio at 100% otherwise. Only the utilization path considers a denominator that is small but not zero.


How Did Two Base Units Become 1.25 Billion BUSD?

The attacker built the denominator by hand. In one transaction at 04:01:12 UTC, wallet kava1uzx5…s255 deposited 3,801.60 BUSD, borrowed 1,267.20 BUSD, and withdrew 1,267.20 BUSD. We queried the module's state at that block height through Kava's REST API. Cash was 6,351.24161944 BUSD, borrows were 1,267.20118686 BUSD, and reserves were 7,618.44280628 BUSD. Add the first two, subtract the third, and net supply is 0.00000002 BUSD. Two base units.

The market's own books made that possible. Our queries for November 1 show the BUSD market carrying about 7,613 BUSD of reserves against about 4,414 BUSD of cash and 10,234 BUSD lent out, with about 449,715 BUSD of deposits on record. Reserves count against supply, so a reserve claim built up over the market's life sat close to the real BUSD left in it, and net supply drifted toward zero as borrowers repaid and withdrew. The attacker only had to tune the remaining gap. We have not reconstructed the November 8 cycles The Block reports, so whether they were rehearsal or positioning is open.

From there the interest model compounded it. With borrows dwarfing net supply, utilization pinned at its 100% cap. The BUSD market's published parameters set a 0.05 base multiplier, an 80% kink, a 5x jump multiplier and a 2.5% reserve factor, which puts the borrow rate at 104% a year at full utilization. Kava compounds that per second, so a block of about six seconds charges ln(2.04) × 6 / 31,536,000 of the principal, about 0.0000136%. On the 1,267.2 BUSD loan our reproduction shows 17,188 base units of interest in the next block, about $0.00017. After the reserve cut, 16,759 units went to suppliers, divided over a net supply of 2. The factor multiplied by about 8,380 in a single block, from 2.762 to 23,147.

After that first block the interest added itself to the supply, so the factor kept climbing by roughly 23,000 per block instead of multiplying. Twenty-nine blocks later, at 04:03:58 UTC, the attacker repaid the BUSD loan and withdrew 7,618.45 BUSD, which stopped the accrual. The factor froze, and it is still there. Kava's API returns a BUSD supply interest factor of 671,186.99 today. Divide by the 2.762 it held before the attack and you get about 243,000, the "roughly 243,000-fold" The Block reported. Wallet A's own BUSD deposit, on our query, went from 5,192.65 BUSD on record before the jump to about 1,261.85 million after it. The Block rounds that to roughly 1.25 billion BUSD of collateral.


How Did the Attacker Get Past the KAVA Borrow Cap?

Kava Lend caps total KAVA borrowing at 448,358 KAVA, per the same parameters, and the module showed 447,640 KAVA already borrowed just before the attack. The cap had about 700 KAVA of headroom. A billion-dollar collateral claim is worth little if the pool will only lend you 700 KAVA.

The attacker got around it with a loop between two wallets, and the liquidation code made the loop free. We read it from the transactions and the module source:

  1. Recycle the BUSD. Wallet A (kava1uzx5…s255) sends real BUSD to wallet B (kava1v2s2…fk4). B deposits it. A withdraws the same BUSD again, which its inflated balance always allows. In the cycle we inspected at height 23485236, this left B with about 60,574 BUSD of recorded deposits, made from about 11,000 BUSD of real tokens sent around repeatedly.
  2. Borrow to the cap. B borrows 445,719 KAVA, almost exactly the cap, at the 50% loan-to-value the market allows.
  3. Liquidate yourself. In the next block, A calls MsgLiquidate on B as the keeper. B keeps the KAVA. The debt is gone, the cap is empty again, and the loop restarts.

Step 3 is the second bug. In the liquidation keeper, when the collateral lot to be auctioned exceeds what the module actually holds, the code shrinks the lot to the module's balance. It still removes the full borrow from the books, and the borrower's record is deleted. Most of the BUSD backing B's deposit had already gone back out to A, so in our reading the auction lot was close to nothing while the full 445,719 KAVA of debt was forgiven.

The first few cycles also emptied the cap's existing usage. Between 04:04 and 04:05 UTC the attacker repaid other borrowers' KAVA loans alongside its first borrows, and in one transaction at height 23484846 B repaid KAVA loans owned by 22 other addresses. B had borrowed about 125,572 KAVA two blocks earlier, so by our reading the attacker was paying off other people's debts with the pool's own KAVA to make room under the cap. That fits Bitquery's note that its net figure accounts for "transactions involving other borrowers."

Our count of B's MsgBorrow messages for KAVA is 127, from the first at 04:04:21 UTC to the last at 05:42:04 UTC. They sum to 55,205,050.6 KAVA. That matches the 127 loans and 55.2 million KAVA Bitquery and The Block report.


The Attack, Step by Step

#Time (UTC, Nov 14)ActorActionEffect
0From Nov 8AttackerRepeated BUSD deposit and borrow cycles (The Block)Preparation in the dormant market
104:01:12Wallet ADeposit 3,801.60, borrow 1,267.20, withdraw 1,267.20 BUSD in one txBUSD net supply set to 2 base units
204:01:18ModuleFirst interest accrual over a supply of 2Supply factor 2.762 to 23,147 in one block
304:01 to 04:03Module29 blocks of accrual at a 104% annual borrow rateFactor reaches 671,186.99, about 243,000x
404:03:58Wallet ARepay BUSD loan, withdraw 7,618.45 BUSDAccrual stops, index frozen
504:04:21 to 05:42:04A and B127 cycles: recycle BUSD into B, B borrows up to 445,719 KAVA, A liquidates B55.2M KAVA borrowed, debt wiped each time
604:04 to 04:05Wallet BDuring the first cycles, repay other borrowers' KAVA loans, 22 in one txKAVA borrow cap emptied
709:59 onwardAttackerAbout 49.2M KAVA bridged to BNB Chain via LayerZero and soldAbout $2.09M in BNB and USDT
811:28KavaFirst public acknowledgment on XMore than seven hours after step 1

Rows 1 to 6 are our reading of the transactions and module state via Kava's REST API; rows 0, 7 and 8 are as reported.


How the Loop Fit Together

rendering diagram…

Which Loss Figure Should You Trust?

Four numbers are circulating, and they measure different things.

FigureWhat it measuresSource
55.2M KAVA, about $3.76MGross KAVA borrowed, valued at the $0.068 pre-attack priceThe Block, Bitquery; our count of 127 borrows sums to 55,205,050.6 KAVA
About 53.9M KAVANet KAVA kept after the loans the attacker repaid for other borrowersBitquery, via FinanceFeeds
48.36M KAVA, about $3.3MKAVA trackers saw bridged to BNB Chain, in early reportingAirdropAlert
About $2.09MWhat the attacker realized selling into a falling marketThe Block

We trust the 55.2 million KAVA gross figure because the chain confirms it. Treat $3.76M as the pre-attack value of what left the pool and $2.09M as what the attacker actually got.

What depositors lost is a separate question. Kava's API still records 55.21 million KAVA of deposits in Kava Lend, while the module account holds about 1,987 KAVA as of November 16, up from the 0.19 KAVA low point after the drain. Kava has not said how, or whether, those claims will be made whole.


Where Did the KAVA Go?

Starting at 4:59 a.m. ET on Saturday, about four hours and seventeen minutes after the last borrow, the attacker bridged about 49.2 million KAVA to BNB Chain through LayerZero. Most of it sold within the hour for 3,006.5 BNB and 231,557 USDT, per The Block. Another 4.58 million KAVA went to five unlabeled addresses on Kava.

Bitquery traced $1.17 million of the proceeds into about 2,190 XMR1 tokens on Hyperliquid across eight accounts, and found 612,416 USDT stopped at the Relay bridge with a BLOCKED_WALLET designation. As FinanceFeeds notes, a blocked transfer still has to be recovered.

Kava paused its Axelar bridge at 9:05 a.m. ET "out of an abundance of caution," The Block reported. The stolen KAVA had left through LayerZero.


What Could Monitoring Have Changed?

Little inside the module, because we found no pause path in the x/hard keeper code. The signals were there all the same. The drain took 127 transactions over 98 minutes, and each step left public chain state behind:

  1. The index jump. The BUSD supply interest factor went from 2.762 to 23,147 in one block at 04:01:18 UTC, three minutes before the first KAVA borrow. A stablecoin market's supply index has no legitimate reason to move 8,380x in six seconds.
  2. The borrow-and-liquidate pair. 127 KAVA borrows of up to 445,719 KAVA, all by one borrower, with liquidations by one keeper address against that same borrower. Two distinct addresses produced the whole pattern while the pool's KAVA fell from 55.33 million to a low of 0.19.
  3. The bridge-out. About 49.2 million KAVA moved through LayerZero starting at 4:59 a.m. ET, per The Block, more than four hours after the last borrow. Kava later paused its Axelar route. Public sources do not say whether it could have paused the LayerZero one.

Either of the first two rules fires while the drain is still running. Neither had anything in the module to pull. A chain-level response has precedent, since Cronos validators halted block production when Tectonic was drained in August, as we covered in the Tectonic post, but that is a validator decision, not a module control.


Why Do Dormant Markets Keep Breaking Lending Protocols?

A lending market whose real liquidity has drained away is still a live set of formulas, and every ratio in it divides by a supply an attacker can set by hand.

We saw a close relative five weeks earlier. BiFi's Ethereum USDT market sat idle for about 36 days, and one caller who controlled the market's totals for a block controlled what that idle interest was worth. Kava Lend adds two conditions: the market's net supply sat near zero, its asset had stopped being minted, and it still held a 50% loan-to-value against everything else in the pool.

The usual comfort is that code which has run for years has been tested by those years. Years of normal use only test the inputs normal users produce, and nobody holding BUSD in Kava Lend was going to set net supply to two base units. The bug in CalculateSupplyInterestFactor was visible in source the whole time, one function away from a guard that got it right.

The liquidation bug made it worse. A liquidation that clears the full debt while the module cannot pay the full lot turns any phantom collateral into permanent bad debt, and gives the attacker an unlimited reset button on the borrow cap.


What Should Lending Operators Do?

  1. Bound every denominator from below. A zero check leaves every tiny positive value open. Refuse to accrue supply interest, or clamp the per-accrual factor, when net supply falls under a minimum in base units, and treat a negative or near-zero cash + borrows - reserves as a halt condition rather than a value.
  2. Cap the per-block change in every index. A supply index that moves more than a small multiple in one accrual is a fault. Revert the accrual instead of writing it.
  3. Retire markets whose asset is retired. BUSD stopped minting in 2023. A market for it should have been set to deposit-only or zero loan-to-value long before November 2026, so that nothing borrowed against it could reach the rest of the pool.
  4. Never forgive more debt than the liquidation recovered. If the lot is clamped to the module's balance, clear only the matching share of the borrow and leave the rest on the books, or fail the liquidation.
  5. Block self-liquidation loops. Flag or reject liquidations where the keeper funded the borrower's collateral within the same window, and alert on repeated borrow-then-liquidate pairs between the same two addresses.
  6. Wire alerts to levers you control. Index jumps, module outflow and bridge deposits should page someone who can pause every bridge route the chain controls.

Would an Audit Have Caught It?

Yes, if the review covered the x/hard interest and liquidation math. Both flaws sit in plain view in the public source: a supply factor that divides by a value guarded only against zero, next to a utilization function in the same file that does bound it, and a liquidation path that clamps the lot but still erases the full debt. The signals above were public chain state, but with no pause path we could find in the module an alert could only start a response, so preventing the drain came down to review. Our DeFi security audits stress-test liquidation mechanisms and the bad-debt scenarios they create, and our L1 chain audits cover Cosmos SDK chains, including the state transition logic where Kava Lend's interest math runs.


Frequently Asked Questions

Was the Kava Lend exploit an infinite mint?

No. Kava said so directly, and the chain agrees: the 55.2 million KAVA the attacker took was existing KAVA borrowed out of Kava Lend's pool, whose balance fell from 55.33 million KAVA to a low of 0.19. What was inflated was an accounting number, the BUSD supply interest factor, which made the attacker's BUSD deposit count as about 1.25 billion BUSD of collateral.

How much did Kava Lend lose?

About 55.2 million KAVA, worth roughly $3.76 million at the $0.068 pre-attack price. Bitquery puts the net amount the attacker kept at about 53.9 million KAVA, and the attacker realized about $2.09 million selling on BNB Chain. Kava has not published an official figure.

What was the root cause of the Kava Lend hack?

By our reading of the source, two flaws in the x/hard module. The supply interest factor divides by net supply and only checks for exactly zero, so two base units of supply inflated it about 243,000x. Liquidation then cleared full debts while auctioning only what the module held, so the attacker could wipe each KAVA loan and reuse the borrow cap.

Has Kava published a post-mortem?

Not yet. As of November 16, 2026, Kava has said it "will share a complete incident report as soon as it is ready," per The Block, and has not given a loss figure or a compensation plan.

Could other lending protocols have the same bug?

Yes. Any lending market that divides interest by net supply and only guards against zero is exposed when that market can be drained to a few base units, which is easiest in dormant markets with little liquidity.


Sources / References

Dmitry Serdyuk
Dmitry Serdyuk

Co-Founder & CDO

Full-Stack Operator | Building across security, AI, and digital infrastructure.