Fluidkey Breach: Logging Tool Enables $1.58M Drain
Originally published at Fluidkey Breach: Logging Tool Enables $1.58M Drain On 19 November 2026, unauthorized transfers left Fluidkey user accounts on Ethereum, Base, BNB Chain, Optimism, Polygon and Arbitrum. Fluidkey t
Originally published at Fluidkey Breach: Logging Tool Enables $1.58M Drain
On 19 November 2026, unauthorized transfers left Fluidkey user accounts on Ethereum, Base, BNB Chain, Optimism, Polygon and Arbitrum. Fluidkey took all services offline at 18:05 UTC and later said a third-party logging tool had exposed session data, allowing an attacker to prepare transfers on Fluidkey's servers that were signed when a user opened the app. Bitquery traced at least $1.58 million to wallets it tied to one attacker between 17:28 and 18:04 UTC. A GitHub tracker recorded $1,634,503. Fluidkey has not published an official loss figure. This is a defensive account of a public incident, not a how-to.
What Happened
Fluidkey pays gas and submits user transactions through a relayer. Bitquery found that relayer never sent more than 30 transactions in an hour on Base and Ethereum on 18 November or the morning of the 19th, then sent 174 in the hour from 17:00 UTC.
At 17:16 UTC, a 5.15 USDC payment left a Fluidkey account on Base toward a wallet that later received gas from the attacker; Bitquery treats this as a test after attacker wallets had been funded from Tornado Cash. The drain began at 17:28 UTC on Ethereum and Base. Bitquery's largest single transaction moved about $415,783 in USDT from two Ethereum accounts, and half of the attacker-tied take had been paid out by 17:33 UTC. The relayer's last transaction in that burst was at 18:04:59 UTC. Fluidkey said it took all services offline at 18:05 UTC and that no unauthorized transaction had gone through since.
Between 20:05 and 21:29 UTC, six wallets deposited 881.6 ETH into Tornado Cash β about $1.57 million at Bitquery's $1,778 per ETH. The tracker, checked at 08:23:08 UTC on 20 November, recorded the same 881.6 ETH and valued those deposits at $1,569,248. At 16:10 UTC on 20 November, Fluidkey said it was back online with v1.79.22, that earlier app versions would stay blocked, and that it would contact affected users and publish a full post-mortem.
How Fluidkey's Accounts Work
Fluidkey gives each payment a new stealth address, so outsiders cannot add up one person's balance. Bitquery, summarizing Fluidkey's technical walkthrough, describes each of those addresses as a Safe smart-contract wallet with a single owner key. The user signs a message with a wallet they already have; application code in the browser derives a spending key, meant to stay on the client, and a viewing key used to generate addresses and notice incoming payments. Dedaub's May 2024 audit of the Stealth Account Kit describes the same outline.
A fresh stealth address holds no ETH for gas, so Fluidkey submits the transaction and pays the fee. Bitquery found that one relayer contract, 0x8090a9dbβ¦058a0, serves all six chains it monitored. We confirmed that this address holds 1,343 bytes of contract code on both Ethereum and Base. Bitquery names the Ethereum signer as 0x7997e874β¦82da and the Base signer as 0x2b40fe2eβ¦854a, both EOAs.
The design concentrates trust: whoever can get the app to sign a prepared payload, and can have the relayer submit it, can move funds out of many stealth Safes without holding those Safes' spending keys.
What the Chain Shows
Bitquery followed every transfer the relayer made to an outside wallet between 17:10 and 18:10 UTC. It counted a wallet as the attacker's only if money or gas tied it to the same cluster. On that standard, $1.58 million reached attacker-tied wallets. Another $65,000 went to wallets with no clear link (Bitquery's upper bound is $1.64 million). A further $349,000 went to wallets with a history, including one user who moved about $277,000; those withdrawals are not part of the $1.58 million.
Attacker-tied payouts, still from Bitquery, were about $837K on Base, $660K on Ethereum, $43K on BNB Chain, $30K on Optimism, $7K on Polygon and $990 on Arbitrum, mostly in USDC and USDT.
The GitHub tracker that Bitquery cites for the $1.63 million figure reported $1,634,503 stolen, with $1,118,222 marked confirmed and the rest including suspected addresses. They classify some wallets differently but agree on the scale and on the 881.6 ETH Tornado Cash deposits. The totals should not be added together, and neither is an official Fluidkey loss total.
Named addresses in Bitquery's record include 0xeb6ed5a9β¦0980 (first payout, $475,967), 0x9e5143a5β¦40ae (largest Base payout, $259,606), and Tornado depositor 0xd3a9d27aβ¦4eea (349.3 ETH). The chain shows that Fluidkey's own relayer sent the payouts; it does not show why.
What Fluidkey Says Caused It
Fluidkey first asked users not to touch the website or app at 18:20 UTC on 19 November, then said at 20:17 UTC that there had been no unauthorized transactions since the 18:05 UTC shutdown, that SEAL 911 was helping, and that services would stay paused. The cause statement came the next afternoon, in the v1.79.22 announcement:
The attacker used flaws in a third-party logging tool to obtain session data, prepare unauthorized transfers on our servers, and have them signed when a user opened the app. Transfers can no longer be prepared this way in v1.79.22, which also includes additional safeguards. The private keys controlling user funds never left the device.
Fluidkey has not named the logging tool or released the promised post-mortem as of this writing.
# Simplified threat-model sketch. Not Fluidkey source, and not a
# reproduction of the incident.
# third-party logging tool --> session data reaches an attacker
# Fluidkey servers prepare a transfer payload
# app signs that payload when the user is next in session
# relayer submits the signed Safe transaction and pays gas
Why "Keys Never Left the Device" Was Not Enough
"Self-custodial" and "the key stays on the device" are often treated as the same guarantee. This incident shows they are not. A signing key that never crosses the network can still authorize a transfer if the application decides what gets signed, and if a server is allowed to stage that payload in a session the app will honor.
That is a different failure than a leaked seed or a compromised Safe module. It is closer to other wallet-adjacent breaks: the Keplr / Web3Auth social-login key exposure moved the custody boundary into a vendor, and the Binance EIP-7702 hot-wallet case showed how a powerful delegation can empty an account without a classic on-chain bug. Fluidkey's official account keeps the spending key on the device, but the session, the server-side transfer queue, and the relayer still sat in the middle of every spend β which is why payouts could run on six chains in the same 36 minutes.
What the 2024 Audit Already Flagged
In May 2024, Dedaub audited the Stealth Account Kit library, not the production app or the relayer. Finding P1 warned that spending and viewing keys are derived from a message signature supplied to browser application code. Legitimate app code, it said, does not store or transmit the spending key. Dedaub warned that a compromised Fluidkey server "could potentially serve malicious javascript code to the user," so "the overall security of stealth addresses currently relies on the security of the Fluidkey server."
That is not a claim that this incident exfiltrated spending keys. Fluidkey says those keys never left the device, and the public evidence is consistent with session-assisted signing of server-prepared transfers, not with a dumped key file. The audit still matters: it recorded, two years earlier, that app and server integrity were part of the trust model, even when the Safe does exactly what its owner signed.
Lessons for Builders
Treat third-party scripts in a wallet app as production credentials. A logging or analytics widget that can see session state belongs in the same threat model as the signer. Isolate or remove such tools from signing surfaces.
Do not let a server prepare an unsigned spend that the next app open will sign. The client should construct the call, the user should see a clear summary, and the server should relay a payload it cannot pre-stage.
A relayer that pays gas for every user is a single on-chain choke point. Rate limits, anomaly detection, and an emergency pause belong there. The jump from at most 30 transactions an hour to 174 in one hour was visible on-chain in real time.
Say clearly what "self-custodial" does and does not cover. The key may be derived in the client, but the client, its session, and any service allowed to propose transactions are still trusted components. The same honesty belongs in account-abstraction and recovery designs and in any review that treats wallet infrastructure as in-scope, including our smart contract security checklist.
What Users Should Do
Follow Fluidkey only through official channels, beginning with @fluidkey and fluidkey.com. The team says it will contact affected users individually and will not reach out privately through social media. Update only through the official app or site to v1.79.22, and wait for Fluidkey's guidance before moving remaining funds. Fluidkey has not released an official loss total, a victim list, or payout transaction hashes; we have not invented hashes the sources did not list.
Sources
- Fluidkey initial incident notice (19 Nov 2026, 18:20 UTC)
- Fluidkey shutdown update (19 Nov 2026, 20:17 UTC): no unauthorized transactions since 18:05 UTC; working with SEAL 911; services paused
- Fluidkey impersonation warning (20 Nov 2026, 08:52 UTC): team will not reach out privately through social media
- Fluidkey v1.79.22 and cause statement (20 Nov 2026, 16:10 UTC): logging-tool session data; server-prepared transfers signed when the app opened; keys never left the device
- Bitquery investigation: $1.58β$1.64 million range; $415,783 largest USDT transaction; relayer contract and signers; 881.6 ETH into Tornado Cash
- Independent drain tracker: $1,634,503 stolen; 881.6 ETH / $1,569,248 into Tornado Cash; checked 20 November 2026, 08:23:08 UTC
- Etherworld summary: Fluidkey timeline and Bitquery's $1.58 million figure
- 0xCERT advisory 0xCERT-2026-0388: Fluidkey's 19 November statements, before the cause was disclosed
- Dedaub, Fluidkey Stealth Account Kit audit (24 May 2024): P1, spending key accessible to application code if a compromised server served malicious JavaScript
- Etherscan relayer and Basescan, same address: 1,343-byte contract code on both chains at our check
Closing Thoughts
The Fluidkey drain did not require a bug in Safe or in the six chains that carried the payouts. A logging tool that could see session data, a server that could prepare transfers, and an app that signed those transfers when a user opened it were enough for the relayer to empty stealth accounts for 36 minutes. "The key never left the device" is important. It is not, by itself, a complete security model for a wallet that still proposes, signs, and relays spends on the user's behalf.
If you are building a smart-account wallet, a relayer, or any client that signs server-assisted transactions, talk to our team or review audit pricing before that path goes live against real user funds.
Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.