We manage your security to weather the AI storm →

← Resource Hub

The Kava Lend Exploit Explained: A Dormant BUSD Market

By Daniel GelfandPublished 10 min read

The Kava Lend Exploit Explained: A Dormant BUSD Market

Quick answer. On November 14, 2026, two wallets borrowed about 55.2 million KAVA from Kava Lend in 127 loans, per Bitquery. A near-empty BUSD market inflated their collateral 243,000-fold. Kava calls it isolated, not an infinite mint, and has given no loss figure.

Watch: The Kava Lend Exploit Explained: A Dormant BUSD Market 1:42
Video transcript

Two wallets borrowed about 55.2 million KAVA from Kava Lend on November 14, 2026. The way in was a nearly empty BUSD market.

The BUSD market held almost no cash. Kava's public code guards only an exactly zero denominator, so the supply index jumped roughly 243,000-fold in three minutes.

Several thousand BUSD then counted as about 1.26 billion. The wallets would borrow KAVA, then usually liquidate the helper wallet, erasing the debt. Room under the cap reopened, and the loop ran 127 times.

These figures sit at different points, so do not add them up. About 53.9 million KAVA was kept, 49.2 million bridged to BNB Chain, and sales returned roughly 2.1 million dollars.

Kava says the incident was isolated, not an infinite mint, and paused the Axelar bridge. The cause is an independent reconstruction. No official loss figure or incident report had appeared by November 16.

Lending protocols should treat a near-zero denominator like zero, bound index moves per update, count caps against cumulative borrowing, and wind down dormant markets.

This is Guardian analysis, not a review of Kava's code. Our public RAAC finding L zero two had a different trigger. In both, supply index growth outran the assets behind it.

Read the full article for the evidence and open questions. Guardian's Managed Security monitors contracts on every block, and an engineer responds when a monitor fires.

What Happened

On November 14, 2026, two wallets drained Kava Lend's KAVA pool through its dormant BUSD market. They borrowed 55,205,051 KAVA in 127 loans between about 04:04 and 05:42 UTC, per Bitquery. Bitquery

Bitquery says the attacker rehearsed with small BUSD deposits, loans and withdrawals from November 8 to 10, and PublicAML traces BUSD conversions on November 6, 8 and 9 into the Kava account that later received the KAVA. PublicAML

Kava acknowledged unexpected activity the same day. It said the incident was isolated to Kava Lend, involved KAVA being bridged and sold, and was not an infinite mint. Kava

Kava said partner exchanges had paused deposits and withdrawals. It later paused the Axelar bridge and promised a complete incident report.

A Dormant Market

Kava Lend is a pooled money market. Lenders supply an asset, and a supply index records how much each deposit has grown. Kava Labs GitHub

Just before the attack, the BUSD market held about 0.003 BUSD in cash. It still owed lenders about 440,145 BUSD. Bitquery

Against that, 5,084 BUSD was out on 37 open loans and 7,618 BUSD sat in reserves, Bitquery found.

A dormant market is not a retired one. It still carries lender claims, reserves, and live code that can value collateral.

The same applies to any old deployment; our guide to legacy smart contract risk covers how to inventory and retire them.

How the Supply Index Broke

Kava's public code computes the supply interest factor as one plus new interest divided by cash plus borrows minus reserves. Kava Labs GitHub

The code returns one only when that denominator is exactly zero. A denominator of a few base units passes through untouched.

Bitquery found the effective BUSD denominator reached two base units. New interest was then divided by almost nothing. Bitquery

The supply index rose from about 2.76 to 671,187 in three minutes, roughly a 243,000-fold jump.

A deposit of several thousand BUSD was then recorded as about 1.26 billion BUSD of collateral.

Kava's last recorded consensus upgrade was v0.28.0, scheduled for February 25, 2025. Public operator data lists v0.28.2 as the current node version. Polkachu

Both tagged releases contain the same exact-zero check. No public record proves which binary every validator ran on November 14.

Index values that jump instead of growing smoothly are a known bug class, covered in our guide to DeFi accounting vulnerabilities.

How 127 Loans Beat a Cap

Kava Lend capped KAVA borrowing at 448,358 KAVA, per Bitquery. The wallets borrowed 55,205,051 KAVA anyway. Bitquery

In nine of ten rounds, the main wallet liquidated the helper one block after the loan, per Bitquery. The liquidation erased the debt, though no real BUSD collateral was available to back it.

With the debt gone, room under the cap reopened. The loop ran 127 times over about 98 minutes.

The KAVA pool held about 55.3 million KAVA before the loop and 519 KAVA right after it. A later November 14 read showed 0.19 KAVA. StalkChain

The wallets also repaid loans other borrowers owed, which freed more room under the cap, Bitquery found. After those repayments they kept about 53.9 million KAVA, taken from the lenders' pool, not new issuance. Kava

A cap that counts only open debt is not a rate limit. If debt can be cleared inside the same flow, the cap resets with it.

Where the KAVA Went

Most of the KAVA left Kava for BNB Chain and was sold there the same morning, Bitquery found. Bitquery

PublicAML reports 49,207,600 KAVA crossed through LayerZero, a route the wallet had tested with a small transfer on November 6. Kava's announced pause covered Axelar, a different route. PublicAML, Kava

Bitquery measured a 36% price fall during the sale. That is why proceeds came in far below the KAVA's pre-attack value. DefiLlama

DefiLlama classifies the incident as Token & Share Accounting, an arithmetic error in Kava Lend's BUSD market.

The figures below sit at different points in the flow, so they should not be added together or swapped.

Bitquery read Kava data to about 16:20 UTC on November 14 and most other chains to about 16:08 UTC. Later transfers or recoveries are not covered.

FigureAmountWhat it measures
Borrowed55,205,051 KAVATotal across 127 loans
RetainedAbout 53.9 million KAVAKept after repaying loans other borrowers owed
RecordedAbout 53.9 million KAVA, about $3.7MDefiLlama hacks record, valued before the selloff
BridgedAbout 49.2 million KAVASent to BNB Chain through LayerZero
SoldAbout 48.4 million KAVASold on BNB Chain in about 35 minutes
ProceedsAbout $2.1 millionReceived after the selling pressure

Confirmed and Still Open

Who absorbs a shortfall depends on a protocol's backstops, which our guide to bad debt in DeFi compares.

  • Kava statement: isolated to Kava Lend, not an infinite mint, exchanges paused, Axelar bridge paused. Kava
  • Onchain evidence: the KAVA pool drain and the move to BNB Chain. Bitquery, PublicAML
  • Reconstruction only: the near-zero denominator, the borrow and liquidate loop, and the earlier rehearsals. Kava Labs GitHub
  • Open: an official loss figure and incident report; neither appeared by November 16. Kava's status page lists no events for November 14 to 16. StalkChain
  • Open: how affected lenders will be treated; no reimbursement plan had been announced by November 15.
  • Open: when the Axelar bridge reopens, and what happens to blocked and remaining proceeds.

Checks for Lending Protocols

Three defects lined up here: a near-zero divisor, a cap that resets, and a market left running. Our lending protocol audit checklist covers the wider set of controls.

  • Treat a near-zero denominator like zero: require a minimum supply before accruing interest to suppliers.
  • Cap how far a supply or borrow index can move per update, and halt the market on a breach.
  • Fuzz the invariant that recorded collateral never exceeds assets the market holds, starting from empty markets.
  • Count borrow caps against cumulative borrowing per period, not only open debt.
  • Flag liquidations that clear debt while seizing little or no real collateral.
  • Wind down dormant markets: settle lender claims, freeze supply and borrowing, then delist.
  • Alert on borrow and liquidation cycles repeating within minutes from the same wallets.
  • Alert when one market's recorded collateral jumps by orders of magnitude, whatever the asset's price.

For Exchanges and Bridges

Exchanges listing an affected token should watch for deposits from the exploit wallets and their bridge destinations, on every chain.

Relay's record shows one order failed with the reason BLOCKED_WALLET, leaving about 612,416 USDT there, per Bitquery. The same record marks the deposit as one the sender can pull back, so the block is not a freeze. Bitquery

A router can refuse a flagged wallet, but only a freeze or seizure keeps the funds.

Bridges should reconcile supply on each side and rate-limit sudden inflows of a token under active incident.

Protocols should keep verified exchange and bridge contacts ready. Our guide on how to request an exchange freeze after a hack lists the evidence to send.

Guardian Analysis: Bound Index Growth

This section is Guardian analysis. It compares a public Guardian finding with the reconstructed Kava mechanism, and does not suggest Guardian reviewed Kava's code.

Guardian's RAAC Core Contracts report includes finding L-02, rated Low and resolved. A permissionless checkpoint path compounded a supply index designed to grow linearly. Guardian

Its proof of concept paid 112,101.56 crvUSD of interest on 100,000 principal under daily checkpoints, against 75,267 with one annual checkpoint.

The extra interest consumed a buffer intended for protocol fees and bad debt.

The triggers differ: call frequency at RAAC, a near-zero divisor at Kava. In both, supply index growth outran the assets behind it.

The check is the same. Bound index growth per update, and test it against independently computed accrual before it changes collateral or credit.

Frequently Asked Questions

Has Kava published a fix for the supply index check?

None is visible publicly. As of November 16, the master branch of interest.go still guards only an exact-zero denominator, and Kava's latest GitHub release is still v0.28.2, a January 23, 2026 dependency update. Whether the code running onchain has changed is not public. Kava Labs GitHub

Where did the rest of the sale proceeds go?

Bitquery traced about $1.17 million to eight Hyperliquid accounts, in a balance snapshot taken at about 16:14 UTC on November 14. Whether any of it has since moved, been frozen or been returned is not public. Bitquery

What should Kava Lend lenders watch for?

Use only Kava's official X account and status page to verify any claim or compensation process. Treat unsolicited recovery or refund offers as likely scams, and never share keys or sign transactions to claim funds. Kava

Sources

Watch Every Market, Including the Quiet Ones

Guardian's Managed Security monitors contracts on every block, including large fund movements, and a Guardian engineer responds when a monitor fires. Talk to us about coverage for your markets.