Dev.to Security 🔐 Cybersecurity 👁 0 📖 9 min read

Chainflip Vault Drain: Valid Aggregate Key Signature

Originally published at Chainflip Vault Drain: Valid Aggregate Key Signature On 21 November 2026, at 07:10:59 UTC, a single Ethereum transaction emptied the Chainflip Vault of every unit of ETH, USDC and USDT it held. I

Originally published at Chainflip Vault Drain: Valid Aggregate Key Signature

On 21 November 2026, at 07:10:59 UTC, a single Ethereum transaction emptied the Chainflip Vault of every unit of ETH, USDC and USDT it held. In block 26,461,896, the Vault at 0xf5e10380…2Bcc sent 520.2243399 ETH, 3,249,631.41 USDC and 758,141.33 USDT to a fresh address. Defimon and Bitquery valued that Ethereum haul at about $4.86 million, with ether near $1,635 at the minute of the drain. Later that day, Chainflip Labs said the same incident also took 122.86137042 BTC from its vaults. This is a defensive account of a public incident, not a how-to.

The Vault did not fail a signature check. The call went through transferBatch, gated by KeyManager.consumeKeyNonce, and Defimon reported that the Schnorr signature verified against the live on-chain aggregate key. The tell was the nonce: a 39-digit number that, in Bitquery's review of every signature the KeyManager has accepted since 2023, was the only one that broke Chainflip's running count.

What Happened

Chainflip swaps native coins across chains. Users deposit into vaults, and validators pay the other side from the matching vault. Official docs describe a FROST threshold scheme: up to 150 validators form the authority set, and 100 of 150 must participate to produce the aggregate signature that can move funds. The current key lives in a KeyManager contract and rotates every few days.

At 07:09:59 UTC, a brand-new EOA, 0x78fa4ac2…2db0, received gas from Tornado Cash. One minute later, in 0xc928950f…f20f, that wallet's first — and only — transaction called the Vault and paid three assets to 0x7d14a719…1Eaf, which had no prior balance.

Defimon Alerts flagged the drain at 07:33 UTC as a compromise of the validator network's aggregate signing key, rather than a signature-verification bug. The last routine rotation, setAggKeyWithAggKey, had occurred about 2.4 days earlier. The Block timed it to 4:53 p.m. ET on 18 November (21:53 UTC), which matches that interval.

Chainflip posted at 09:14 UTC that it was aware of a potential exploit affecting its vaults and was putting the protocol into full safe mode, disabling deposits, withdrawals, swaps and quoting. Bitquery found that ordinary Ethereum payouts continued for almost two hours after the drain — 63 of them, still under the same key — until validators moved to a new key at 08:53:35 UTC, about 13 hours early. Bitquery recorded no further Ethereum signatures after that switch, at a check at 11:23 UTC.

How the Vault Authorizes Transfers

Ethereum cannot verify FROST Schnorr signatures natively, so KeyManager stores the aggregate public key and checks every payload that would move funds. Official docs also explain transferBatch: one signature can cover several transfers, which saves gas. The same design is documented for other EVM deployments, including Arbitrum, BNB Chain and Tron. A key rotation does not migrate balances. It updates the key in KeyManager; payloads signed by the previous key then fail, while the Vault and its deposit channels stay in place.

// Simplified illustration of the authorization pattern in Chainflip's
// public vault docs. Not verbatim source, and not a reproduction of
// the incident.
function transferBatch(SigData calldata sigData, TransferParams[] calldata transfers) external {
    // KeyManager checks the FROST/Schnorr signature against the live
    // aggregate key and consumes sigData.nonce so it cannot be replayed.
    keyManager.consumeKeyNonce(sigData, /* message hash of this call */);
    for (uint256 i = 0; i < transfers.length; i++) {
        _transfer(transfers[i].token, transfers[i].recipient, transfers[i].amount);
    }
}

That pattern holds only if the aggregate key stays under an honest threshold of validators, and if signatures come from the protocol's normal signing path. A valid signature is not, by itself, proof that the request was an authorized protocol payout.

What the Ethereum Transaction Shows

We re-read the drain and the Vault's balances on a public Ethereum RPC. In the block before 26,461,896, the Vault held 520.22433990 ETH, 3,249,631.410561 USDC and 758,141.332716 USDT. In the block of the transaction, those balances were zero, and the recipient held the matching amounts. That agrees with Chainflip's official figures to the precision the team published, and with Defimon's and Bitquery's reads of the same transfers.

The selector is 0xdf5ed9ea (transferBatch). It encodes three transfers to the same recipient: native ETH, as Chainflip's 0xeee…eee sentinel, plus USDC and USDT. A KeyManager log at the official address 0xcd351d36…f08Be was emitted in the same transaction, consistent with consumeKeyNonce accepting the payload. There is no failed check, forced storage write, or unexpected external call in this transaction. The contracts did what a valid aggregate signature is meant to authorize.

The Nonce That Broke the Count

Each Chainflip Ethereum signature carries a nonce that KeyManager records so an old payload cannot be replayed. Bitquery says the protocol normally draws those nonces from a running count, with occasional gaps of about one number in twenty that never appear on Ethereum.

The drain did not use the next integer. We decoded the nonce in the payload as 190262553000735942938746936035352881278, a 39-digit value. Bitquery, reading all 299,677 KeyManager signature logs since 2023, found this was the only accepted nonce off the running count. Around the drain, the ordinary count stood at 314,568 before and 314,569 after. The off-count nonce arrived in between two otherwise ordinary signatures.

That does not, by itself, explain how the signature was made. Bitquery is careful: a number off the count means the payment did not come through the usual process. A compromised key would explain it; so would enough stolen shares, or a verification flaw that let a bad signature pass. Defimon leaned toward key compromise because the signature verified against the live aggregate key and the last epoch rotation had looked normal. Chainflip later said it had "identified the mechanism of key exposure," which is an official statement that key material was exposed, not a published proof of the exact mechanism.

Where the Ethereum Funds Went

PeckShield, Lookonchain, Bitquery and CryptoWatchDesk describe the same route. At 07:12:59 UTC, the receiver swapped the USDC on Uniswap in 0xca19dc5f…57d70. The USDT followed at 07:14:23 UTC. Proceeds went to 0xfd24197e…3B080, which also received the Vault ETH.

By 07:19 UTC, that collector had deposited 2,909 ETH into Tornado Cash. Bitquery counted 38 deposits in three batches over 48 seconds: 29 of 100 ETH and 9 of 1 ETH. PeckShield reported the same 2,909 ETH figure. The Ethereum trail, from gas funding to the last mixer deposit, took under 10 minutes. Chainflip's later return offer of 2,618.1 ETH is 90% of that 2,909 ETH total.

What Chainflip Has Said

The 09:14 UTC post confirmed awareness and safe mode but gave no loss figure or cause. In the evening, Chainflip said it had messaged the address holding the BTC taken on 21 November, and published terms at chainflip.io/return:

On 21 November 2026 you took 122.86137042 BTC, 520.2243399 ETH, 3,249,631.41 USDC and 758,141.33 USDT from the Chainflip vaults.

The offer is to return 90% and keep 10% as a bounty if 110.57523337 BTC and 2,618.1 ETH arrive by 18:38 UTC on 23 November 2026. Chainflip says it reported the theft to law enforcement, flagged the addresses, and traced Tornado Cash funding on 11 November, a FLIP purchase, a validator account registered on 20 November, and the server behind that account. It writes that it has identified the key-exposure mechanism and prevented such future attacks, but it has not published a technical post-mortem, Bitcoin transaction identifiers, or a reimbursement plan as of this writing.

CryptoWatchDesk put the combined take at about $12 million after the official BTC figure appeared. We have not independently priced the 122.86 BTC, and Chainflip has not published a single official dollar total, so the hard figures remain the official asset amounts plus the widely cited $4.86 million Ethereum-vault valuation. Bitquery, writing before that BTC figure, had seen no spending from the Bitcoin vault address it watched since Friday and did not check Solana; we have not invented a Bitcoin hash. Its morning check of the Arbitrum vault found ordinary-looking payouts, the largest about $50,000 in USDC — a snapshot, not proof that every non-Ethereum vault was untouched.

On 12 September 2026, an attacker abused how the protocol processed Tron memos and took about $736,000 in USDT. Validators later voted to reimburse affected liquidity providers. That was a chain-specific processing bug. Saturday's drain is a different failure: the Vault accepted a signature that checked out.

Lessons for Builders

A passing signature is not the same as an authorized protocol action. A vault that pays whoever presents a valid threshold signature will do exactly that, including for a payload built outside the normal signing pipeline. Detection has to sit around the key ceremony and the nonce allocator, not only inside the verification check. The same lesson appears when a leaked or delegated signer empties an otherwise correct contract, as in the Keplr / Web3Auth social-login key exposure and the Binance EIP-7702 hot-wallet case.

Sequence anomalies are a real-time signal. One 39-digit nonce among 299,677 otherwise sequential signatures would have been visible the moment it landed. Alerting on unexpected nonces — and on transfers that empty a vault to a brand-new recipient — belongs next to ordinary balance monitoring.

Threshold signatures still concentrate operational risk. A 100-of-150 FROST key is not a single hot wallet, but the ceremony that generates it, the machines that hold shares, and the process that admits a new validator are all in the trusted computing base. A validator registered the day before a drain does not prove that validator was the cause. It is a reminder that onboarding and share hygiene belong in the same review as the Solidity that consumes the nonce.

Rotation speed matters once a bad payload has landed. Ordinary payouts continued under the same key for almost two hours after the Ethereum vault was emptied. An emergency rotate-and-pause path should not wait for a scheduled epoch, and the same authority set signs for several chains, so a compromised key remains a problem until every chain's KeyManager is updated. For a pre-launch pass over vaults, signers and operational keys, use a smart contract security checklist that treats key management as in-scope.

What Users Should Do

Follow Chainflip only through official channels, beginning with @Chainflip and chainflip.io. Do not send funds to a Chainflip deposit address, and do not route swaps through wallets that use Chainflip, until the team says the protocol is out of safe mode. If a swap was already in progress, keep the transaction hash and wait for official instructions. Ignore direct messages that offer refunds or "recovery."

Sources

  • Chainflip safe-mode notice (21 Nov 2026, 09:14 UTC): potential exploit affecting vaults; deposits, withdrawals, swaps and quoting disabled
  • Chainflip bounty and BTC confirmation (21 Nov 2026, 19:22 UTC): on-chain message to the BTC holder; 10% bounty if 90% is returned by 18:38 UTC on 23 November
  • Chainflip Labs, return terms: official amounts of 122.86137042 BTC, 520.2243399 ETH, 3,249,631.41 USDC and 758,141.33 USDT; statement that the key-exposure mechanism has been identified
  • Defimon Alerts (21 Nov 2026, 07:33 UTC): $4.86M Ethereum vault drain; Vault.transferBatch / KeyManager.consumeKeyNonce; signature verified against the live aggregate key
  • Bitquery investigation: $4.86M at $1,635/ETH; 39-digit nonce uniquely off-count among 299,677 signatures; 2,909 ETH into Tornado Cash by 07:19 UTC; key rotation at 08:53:35 UTC
  • PeckShieldAlert: stablecoins swapped for 2,909 ETH, then deposited into Tornado Cash
  • Lookonchain: $4.86M Ethereum composition; funds swapped and deposited into Tornado Cash
  • The Block: 100-of-150 signing, last regular rotation at 4:53 p.m. ET on 18 November, FLIP down about 34%, prior 12 September Tron loss of about $736,000
  • CryptoWatchDesk: Ethereum timeline and addresses; later combined estimate of about $12 million after the official BTC figure
  • Chainflip, EVM vault design and FROST signature scheme: KeyManager verification, transferBatch rationale, 100-of-150 threshold
  • Chainflip mainnet addresses: Ethereum Vault and KeyManager addresses used above
  • Etherscan, drain transaction: transferBatch at 07:10:59 UTC in block 26,461,896; independently confirmed against a public RPC

Closing Thoughts

The Chainflip Vault did not need a missing access check to lose its Ethereum balance. It needed a signature that its own KeyManager would accept. That is the point of a threshold-signed vault, and it is also the point of failure when the aggregate key, or the process that produces signatures, leaves the honest validator set. The off-count nonce is the on-chain clue that this payload did not come from the usual pipeline. The official return message is the source for the Bitcoin amount, and for the team's claim that the exposure path has been closed.

If you are shipping a cross-chain vault, a threshold signer, or any contract that treats a valid aggregate signature as sufficient authorization, talk to our team or review audit pricing before that key can move customer funds.

📰 Read the original article on Dev.to Security

Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.