Users of Fluidkey, a self-custodial privacy wallet built on Ethereum stealth addresses, lost at least $1.58 million on Thursday 19 November 2026, between 17:28 and 18:04 UTC, according to an on-chain investigation by the data firm Bitquery. By Bitquery’s reading of the transfers, the payouts left user accounts on six blockchains through Fluidkey’s own relayer, and half of the attacker’s take was out within five minutes.
The company gave its account of the Fluidkey hack on Friday 20 November at 16:10 UTC, in a post on X announcing that the wallet was back online as version 1.79.22. It said the attacker exploited flaws in a third-party logging tool to obtain session data, then used that data to “prepare unauthorized transfers on our servers, and have them signed when a user opened the app”. In the same post it wrote: “The private keys controlling user funds never left the device.”
That explanation is Fluidkey’s alone. No independent post-mortem exists and the logging tool is not named. At 05:55 UTC on Saturday 21 November, the relaunch post was still the latest on Fluidkey’s X account, and neither that account nor the company’s blog carried a loss figure, a count of affected users or a compensation plan. By the evening of the theft, six wallets tied to the attacker had put 881.6 ETH into Tornado Cash, Bitquery found, and the trail stops there.
Fluidkey hack timeline: 36 minutes on Thursday evening
Bitquery’s reconstruction starts in the early afternoon of 19 November, when a new wallet received its first ETH from Tornado Cash. A second wallet was funded the same way soon after. Both sent small amounts of ETH to other chains to pay for gas.
At 17:16 UTC the first wallet sent tiny ETH deposits to a Fluidkey account on Base. In the same minute, that account paid 5.15 USDC to a wallet that later got its gas from the attacker, a move Bitquery reads as a test. Ten minutes later, by Bitquery’s account, the same wallet sent tiny deposits to two more accounts, and the relayer then drained all three. The firm takes the deposits as a sign that the targets were picked in advance.
The drain proper began at 17:28 UTC, on Ethereum and Base within seconds of each other. The largest single transaction took $415,783 in USDT from two accounts on Ethereum. By 17:33, half of what the attacker would collect had been paid out. The relayer’s last transaction in the burst is stamped 18:04:59. Its activity gives a measure of the hour: on 18 November and the morning of 19 November, the relayer never sent more than 30 transactions an hour on Base and Ethereum combined, Bitquery says. Between 17:00 and 18:00 UTC on Thursday it sent 174.
By Fluidkey’s account, all its services went offline at 18:05 UTC. Its first public word came at 18:20 UTC, in a two-sentence post on X: “We recommend not interacting with the Fluidkey website and app while we investigate the incident.” The notice spoke of a security incident affecting users and did not say whether funds had been lost.
A second post at 20:17 UTC said “there have been no unauthorized transactions since 18:05 UTC, when we took all services offline”, that the team was working with SEAL 911, the round-the-clock incident-response service of the Security Alliance, on the investigation and on tracing the funds, and that the web app and all APIs would stay paused during the review. Bitquery looked at the relayer again at 07:35 UTC on Friday 20 November and found nothing had gone through it on any of the six chains since the last transaction of the burst.
Where the $1.58 million went, chain by chain
Bitquery read every payout the relayer made between 17:10 and 18:10 UTC on 19 November and sorted the receiving wallets into three groups: wallets tied to the attacker by money or gas flows, wallets with an earlier history that look like the users’ own, and wallets with no link either way. The first group received at least $1.58 million. Going by Bitquery’s table, Base and Ethereum account for about 95% of it.
| Chain | Paid to attacker-tied wallets (Bitquery) |
|---|---|
| Base | about $837,000 |
| Ethereum | $660,000 |
| BNB Chain | $43,000 |
| Optimism | $30,000 |
| Polygon | $7,000 |
| Arbitrum | $990 |
In Bitquery’s list of wallets, the first payout address received $475,967 and the largest payout address on Base $259,606.
Bitquery gives two figures for addresses with a history, and they do not measure the same thing. The $349,000 in its payouts section covers the whole hour it read, from 17:10 to 18:10 UTC, and includes ordinary payments made before the drain began; the largest of those went to an exchange deposit address minutes before the first theft. The $292,000 in its key findings counts only what went to such wallets while the drain was running. Of that, about $277,000 was a single user moving funds across Ethereum and Base to a wallet in use since 2024; the money was still there when Bitquery checked, and the firm treats it as the owner’s own move. Another $65,000 went to wallets Bitquery could not place on either side, most of them new. Count those as stolen and the total rises to $1.64 million, the top of Bitquery’s range.
An independent tracker published on GitHub puts the loss at $1.63 million, inside that range, but it gets there differently. It marks $516,000 as suspected rather than confirmed, and Bitquery says at least $229,000 of that sum is the attacker’s by its own reading. The two also disagree on one Polygon payout. Tokenised shares held on BNB Chain, worth about $13,500 at the day’s prices, were moved as well, and $5,500 of them reached the attacker’s wallets. The tracker’s total leaves them out, Bitquery notes.
881.6 ETH into Tornado Cash in under 90 minutes
The stolen tokens were swapped to ETH and collected on Ethereum, Bitquery’s tracing shows. By its count, six wallets made 59 deposits into Tornado Cash’s router between 20:05 and 21:29 UTC on Thursday, for 881.6 ETH in all, or about $1.57 million at the $1,778 per ETH that Bitquery used. The largest of the six put in 349.3 ETH; the others 170.3, 134, 104, 91 and 33 ETH.
Etherscan’s page for the largest depositor confirms the figure for that one wallet: 19 deposits on 19 November in the mixer’s fixed sizes of 100, 10, 1 and 0.1 ETH, adding up to 349.3 ETH. The 881.6 ETH total for the six wallets is Bitquery’s sum. A mixer pools deposits and pays them out to new addresses, so the link between sender and receiver is lost on the public ledger. Neither Fluidkey nor Bitquery reports any freeze or recovery. Tornado Cash’s legal status, and the liability of the developers who wrote it, are covered in our guide to DeFi regulation in the EU and the US.

What Fluidkey says happened, and what the chain cannot show
Bitquery’s report was written before Fluidkey gave a cause, and it is careful about what transfers can prove. The payouts went through the relayer, but the data cannot tell apart a break-in on Fluidkey’s servers, a flaw in the relayer contract or a stolen key on the company’s side. Bitquery adds that it has not read the contract code and that the attacker is unidentified: the wallets are grouped by the money and gas that passed between them, nothing more. “Fluidkey’s own report should settle it,” the firm writes.
Fluidkey’s statement on Friday points to a fourth route. In its telling, nobody stole a key and nobody broke the contract. Flaws in a logging tool supplied by a third party gave the attacker session data. In web services that term usually means the information that identifies a logged-in user to the server; Fluidkey does not say what its own contained. With it, the company says, the attacker prepared transfers on Fluidkey’s servers, and a user’s own app signed them when it was opened.
If that account holds, no key was copied, and the app still signed transfers its user never asked for, because the server presented them. No outside party has checked this version, and the promised post-mortem has not been published. Fluidkey says transfers “can no longer be prepared this way” in version 1.79.22, which it says also adds further safeguards. It does not say which tool was involved, for how long session data was exposed, or whether a user who never opened the app during the 36 minutes could have lost money.
Stealth addresses, Safe accounts and the relayer
Fluidkey’s promise is privacy for incoming payments. Instead of one public address that shows everything a person has received, each payment lands on a fresh stealth address that outsiders cannot connect to the others. According to Fluidkey’s technical documentation, the user signs a message with an existing Ethereum account and two keys are derived from that signature on the user’s device: a spending key that controls the funds and a viewing key that finds them.
The documentation says the key pair “never leaves the user’s client”. A derived branch of the viewing key is shared with Fluidkey, so its server can watch the chain and generate addresses. Each stealth address is a Safe smart account with a single owner, and the contract is deployed only when the user first withdraws from it.
The relayer is the piece that carries transactions to the chain, and the description of it here is Bitquery’s. A fresh stealth address holds no ETH to pay network fees, so Fluidkey sends the transaction itself and pays the gas: on each chain, the firm writes, one Fluidkey wallet signs and sends through one smart contract, the same contract on all six chains. Every payout Bitquery traced passed through it. That marks a limit of self-custody in an app built this way: the keys sit with the user, while, on Fluidkey’s own account of Thursday, transfers could be prepared on its servers for the app to sign.
A 2024 audit flagged trust in Fluidkey’s server, not this attack
Bitquery cites an audit by the security firm Dedaub dated 24 May 2024 among its possible explanations. The audit covered Fluidkey’s open-source Stealth Account Kit, a TypeScript library, and says so plainly: it “covers the open source library, and not the Fluidkey system itself”. Two auditors spent two and a half days on it. It lists no critical or high-severity finding, and nothing in it says Fluidkey left a reported flaw unfixed.
One item, labelled P1, rated “info” and filed as a protocol-level consideration, notes that the private spending key is available to application code running in the browser. Dedaub’s warning was that if the Fluidkey server were compromised, “it could potentially serve malicious javascript code to the user”, and it concluded that the security of stealth addresses “currently relies on the security of the Fluidkey server”.
That is a design note about how much a user of any browser-based wallet of this kind must trust the server. It is not a description of this incident, and Bitquery, which raised it, lists it as one possibility among three. The audit does not mention a relayer, session data, logging tools or prepared transfers, and nobody, Fluidkey included, has said that malicious code was served to users on Thursday.
What Fluidkey is telling users
Fluidkey’s instructions so far fit in two posts. The relaunch post says the wallet is back on web, iOS and Android as version 1.79.22, that earlier versions “will remain offline”, and that all of them are blocked from transacting. To send funds again, users are told to open the app and go to Settings, then Update.
The second is a warning about impostors, posted at 08:52 UTC on Friday 20 November: “Our team will not reach out to you privately through social media.” Anyone in doubt is told to contact support through the company’s website. In the relaunch post, Fluidkey also says it will be “reaching out to affected users individually”; the Thursday update named its channels as its X account and “individual support chats”.

The company has promised a full post-mortem and has given no date for it.
The same pattern as Bloom: keys intact, money gone
This is the second time in a month that a wallet service has said its keys were safe while users’ balances left. On 27 October, the Bloom trading bot lost 9,408 SOL in 94 seconds; Bloom said no keys had leaked, and Bitquery found that each Bloom wallet had signed its own drain. On 11 November, the bridge Across lost control of its web domain, and Bitquery put the losses of Across users near $402,000. Across promised to repay them. Fluidkey has made no such commitment.
What is still missing
Fluidkey has not named the logging tool or said how its flaws were found. It has not published a loss figure, so the only numbers are Bitquery’s $1.58 million to $1.64 million and the independent tracker’s $1.63 million. It has not said how many accounts were emptied, nor whether it will reimburse them.
Bitquery leaves two questions open of its own. The independent tracker flags a possible earlier run about five hours before the drain, with roughly 60% confidence; Bitquery saw the same pattern, four new wallets each receiving about $4,500, and left them out because nothing ties them to the attacker. And $65,000 of payouts sit in wallets that could belong to either side.
The next document to watch is Fluidkey’s post-mortem. Until it appears, the cause of the drain rests on one paragraph in a post on X, published 22 hours after the last stolen transfer.
