Mercado Bitcoin Hack: On-Chain Findings
Back to Insights
BLOG

Mercado Bitcoin Hack: On-Chain Findings

Greg Schneider
Published

Two weeks later, most stolen funds remain in wallets linked to the attacker

In short

  • On 2 November 2026, attackers drained the hot wallets of Mercado Bitcoin, Brazil's largest crypto exchange, in under half an hour. Loss estimates run from about US$38 million (the exchange) to US$48.6 million (public analysis).
  • Chainlabs maintains a free, labelled dataset of the wallets linked to the theft. It now covers 17 chains.
  • More than two weeks on, most of the money has not been cashed out. On 16 November, 10,182 ETH and 113 BTC sat in attacker wallets, and about US$9.5 million of stablecoins remained frozen.
  • The attacker prepared the evening before the theft, has since converted most of it towards ETH and Bitcoin, lost most of what it put into leveraged trading and sent only small test amounts on.
  • No large deposit has yet reached an exchange or OTC desk. That deposit is where an on-chain trail meets a customer record, and it is what labelled wallet data exists to catch.

What happened at Mercado Bitcoin?

Mercado Bitcoin suspended deposits and withdrawals on Monday 2 November 2026 after its hot wallets were emptied on at least ten blockchains. The exchange said the incident was contained, that no customer was harmed and that no data was breached, and it published the destination addresses so others could follow the funds (CoinDesk).

The transfers were ordinary signed transactions, not a smart contract exploit, and how the attacker obtained signing access has not been disclosed (CoinDesk). No one has been publicly identified as responsible, and there is no public evidence tying the theft to the Binance hack a few days earlier.

The size of the loss depends on who is counting. The exchange has put it at about US$38 million, while one public analysis values the assets at US$48.6 million at the time of transfer, across 11 chains (public analysis).

Our own read of the chains matches the public timeline and adds detail to it:

When (UTC)What happened
1 Nov, 18:22The attacker's Ethereum address receives its first ETH. Its Tron address receives its first USDT four minutes later. Both are tested with small payments for two hours.
2 Nov, 06:24 to 06:48The Ethereum hot wallet pays the attacker 2,852 ETH and about 190 token balances. Hot wallets on Bitcoin, Solana, Stellar, Tron, Arbitrum and Avalanche are emptied in the same window.
2 Nov, 06:41102.03 BTC leaves the Bitcoin hot wallet in one transaction.
2 Nov, from 12:00More than 36,000 customer deposit addresses on Ethereum each send their balance to the attacker.
2 to 4 NovTether, Circle and Paxos freeze stablecoins and tokenised gold at attacker addresses.

The deposit address sweep is the detail that matters most for the exchange's customers. It continued for hours after the hot wallets were drained, which points to the attacker being able to sign for deposit addresses as well as hot wallets. The public analysis reaches the same view and advises against reusing old deposit addresses until the exchange confirms they are safe. Read the public analysis.

What did the attacker do before the theft?

The attacker's addresses were funded and tested the evening before. Nine wallets took part. Most sent small test payments to the attacker's Ethereum and Tron addresses between 18:21 and 20:30 UTC on 1 November, ten to twelve hours before the drain, and the rest had funded those wallets beforehand.

These wallets never received stolen funds, so a list of "wallets that hold stolen money" misses them entirely. They are in the dataset as pre-hack funders for two reasons.

  • They were not all new. Some had been active for weeks or months, and one had been dormant since February 2021 before it sent the attacker's Ethereum address its first ETH.
  • Some were themselves funded from the hot wallets of large centralised exchanges. An exchange withdrawal has an account behind it, and an account has a customer record. That makes these wallets a route to a person that does not depend on where the stolen money goes next.

If the same wallets fund the preparation of another theft, anyone holding this list can make the connection on day one.

Where did the stolen funds go?

The attacker has turned most of more than 200 different assets on 17 chains into two: ETH and Bitcoin. About 2 million ARB and 6.7 million XLM were never converted. Almost every step ran through a bridge or a swap service that takes no account.

DateMovementThroughAmount
2 NovSolana to EthereumMayan742 ETH received
3 NovTron to EthereumBridgers (SWFT)1,708,687 TRX in two orders
3 to 4 NovTokens to ETH on EthereumMetaMask swaps, from dozens of fresh walletsMost of the stolen tokens
4 NovEthereum to BitcoinMetaMask Bridge, Rango, Chainflip18 swaps, about US$399,000, 6.13 BTC
4 NovEthereum to BNB ChainAcross42.5 ETH
4 NovTron to other chainsHiFiSwap and Chainflip441,726 TRX and 40,000 USDT
4 to 5 NovXDC, Mantle, Cronos, Linea, Optimism and Polygon to EthereumOne new address that sold and bridged the balances3.07 million XDC, 15,818 MNT, 29,245 CRO and smaller sums
7 NovAvalanche to EthereumMetaMask Bridge71,951 AVAX out, 261.85 ETH in within the hour

Two patterns stand out. The attacker splits work across fresh wallets, each funded with a little ETH for fees, used once and emptied back into a handful of holding wallets. And the attacker returns to the same services, which means each of those services holds a set of order records for this one actor.

Where are the funds now?

Most of the money is still sitting in wallets the attacker controls. Balances below were read on 16 November 2026 at 22:08 UTC.

ChainHoldingsState
Ethereum, Arbitrum and Base10,182 ETH in all: 9,640 on Ethereum, where ten wallets hold more than 100 ETH each, 516.65 on Arbitrum and 25 on BaseEthereum is up from 9,318 ETH on 4 November as other chains were bridged in. The Arbitrum and Base ETH was already there on 4 November but missing from our earlier lists
Bitcoin106.07 BTC at the first attacker address; 6.93 BTC at a consolidation addressThe 106 BTC has never moved
Stellar6,712,500 XLMNothing has moved
Tron69,528 TRX gathered in one new wallet708,244 USDT frozen at the main address
Cronos17,772 CROPartly moved on 9 November
Frozen by issuersAbout US$9.5 million of dollar stablecoins, 368,696 EURC and 369.8 PAXGAll still frozen on 16 November, read again from the token contracts. Two small frozen amounts on Arbitrum and Optimism, about 11,000 USDC, were missing from our earlier lists

The public analysis puts total freezes at about US$11.8 million across five issuers. The difference is mostly in the counting: that total values the euro stablecoins and tokenised gold in US dollars, and adds two smaller tokens we could not confirm with a standard contract call (public analysis).

What has changed since the first week?

Four things. Three of them bring the attacker closer to a party that keeps records; the fourth is money that has not moved at all.

Three attacker orders have not paid out. The public order record of Bridgers shows two orders signed by attacker wallets, 280 ETH for Bitcoin on 4 November and 427,180 TRX for ETH on 3 November, with a waiting status. On chain, the payout addresses received nothing for these orders and the senders received no refund. The Relay bridge's public record shows a third: 17,274 USDC that an attacker wallet sent on 3 November for a transfer to Hyperliquid, marked failed with the reason "BLOCKED_WALLET", with no payout and no refund. We report each status as the service words it; what the services did with the money is theirs to say.

The leveraged trading lost most of the money put in. Between 4 and 10 November, 199,458 USDC was credited to three attacker addresses on Hyperliquid, a derivatives exchange with public trading data. By 16 November no position was open on any of the three accounts. The largest is empty: its closed trades lost US$102,736 and cost US$46,517 in fees, and its last 7,715 USDC went back to an attacker wallet on Ethereum on 13 November.

Small cash-outs have started. Since 7 November the attacker has sent modest amounts through one-time wallets into deposit addresses that other customers also use: 13 ETH into one address, 5 ETH each into two more and about 11,000 USDT into the deposit account of a large service. Separately, 11,473 CRO went to another wallet, where they still sit. These are test-sized, but they are the first transfers to land where a customer file is likely to exist.

Some of the money is parked. Four wallets that have never sent a transaction hold about 2 million ARB, 25 ETH, 21,860 USDT and 14,863 USDC, received from attacker wallets between 2 and 7 November. Who holds their keys is not known.

What has not happened yet is a large deposit at an exchange or OTC desk. On 4 November we checked every attacker address against the Chainlabs label base and found no such deposit. The stolen funds were in self-custody wallets, bridge contracts and swap services.

Why does labelled wallet data matter in a hack like this?

A list of attacker addresses tells you what to block. It does not tell you who is on the other side when the money finally lands. That takes labels: knowing that a given address belongs to a specific exchange, OTC desk, payment processor or swap service.

The Mercado Bitcoin case shows the difference at three points.

  • At the source. Chainlabs already labelled Mercado Bitcoin's hot wallets on Solana and Stellar before the theft, so the drained wallets on those chains were identified as the exchange's own from the first read. The public analysis we reviewed did not name them.
  • Before the theft. The pre-hack funder wallets only become leads once you can say which exchange's hot wallet funded them.
  • At the exit. The attacker holds about US$18 million of ETH and has started testing deposit addresses. When a larger deposit comes, it may well go to a regional exchange or OTC desk that global lists cover thinly. Whoever recognises that address first can act while the funds are still there.

The exit is the part that favours local coverage. Chainlabs labels the wallets of more than 50 OTC desks and local exchanges in Brazil, within a label base of more than 800 regional exchanges and 2,000 OTC desks worldwide. Those are the venues a screening team needs named when funds from a Brazilian exchange hack start to leave.

Speed matters as much as coverage. In one week this attacker opened eight new wallets across ten chains and passed funds through eight more. A list downloaded on day three was out of date by day five.

What is in the Chainlabs Mercado Bitcoin dataset?

The dataset is one free file of wallets linked to the theft, each with its chain and one of three categories. On 18 November 2026 it held 101 wallets across 17 chains (130 rows, one per wallet on each chain).

CategoryWalletsWhat it meansWhat to do with it
Hacker84Linked to the attacker. Received or moved stolen funds.Block
Pre-hack funder9Funded or tested the attacker's addresses before the theft. Never received stolen funds.Watch, do not block on this alone
Victim8Mercado Bitcoin's own wallets that the funds were taken from.Do not block

A wallet is labelled hacker only when the evidence shows the attacker controls it, not merely because the attacker paid it. Attackers send funds to addresses they do not control to mislead whoever follows them, and look-alike tokens and poisoning addresses crowd around every large theft. Those are kept out.

That rule has already cost us rows, which is how it should work. The second review removed one wallet whose only link was a zero-value token transfer. A later check removed an address that turned out to be a contract, not a wallet. Each correction is recorded.

The list also grows. In the week after launch the attacker opened eight new holding wallets and swap payout addresses on Ethereum, Tron, Bitcoin, Avalanche and six smaller chains, and all eight have been added. A full read of Arbitrum, BNB Chain, Base and Optimism on 16 November found five more attacker wallets and four known attacker wallets on six more chain rows, and all of them have been added. Eighteen more wallets received attacker funds: fourteen forwarded them on, and four have held them without ever sending a transaction. For those the money is proven but the key holder is not, so they are held back from the public list until it is.

How can you check an address against the dataset?

The dataset is free and needs no account. There are two ways to use it, and both serve the same current file from chainlabs.ai/mercado-bitcoin-hack-2026.

  • Download the CSV. Three columns: wallet, chain, category. Use it for bulk checks and for loading the hacker rows into deposit screening. The download link stays the same when the file changes, so it can be pulled on a schedule.
  • Connect it to your AI agent. Add the Chainlabs Mercado Bitcoin connector to Claude or any MCP-compatible AI agent, then ask in plain language: "Is this address linked to the Mercado Bitcoin hack?" or "List the hacker addresses on Tron." The AI agent checks against the latest version every time, which matters when the attacker opens new wallets daily.

One limit applies to both. An address that is not on the list has not been linked to this hack by Chainlabs. That does not mean it is safe, and the dataset covers this incident only.

Who is Chainlabs?

Chainlabs identifies who is behind blockchain addresses and licenses that data to compliance, analytics and investigation teams. The Chainlabs label base covers more than 510 million addresses across more than 30 blockchains, and an estimated 90% of the KYT market runs on Chainlabs data. Every attribution is supported by verifiable source evidence, available on request.

The Mercado Bitcoin dataset is a small, free slice of that. The Chainlabs MCP server, now in development, brings the wider label base into Claude or any MCP-compatible AI agent, so that the AI agent works from trustworthy context instead of guessing who an address belongs to. Each attribution comes with a confidence level of High, Medium or Low. It starts with Bitcoin, with Ethereum to follow, and access is by application.

To license the Chainlabs dataset or apply to test the Chainlabs MCP server, email mcp@chainlabs.ai.

Written by Greg Schneider

Sources

Chainlabs figures come from Chainlabs's own reads of public blockchain data between 4 and 16 November 2026 and from the Chainlabs Mercado Bitcoin dataset as it stood on 18 November 2026.

External sources: