Skip to content

exploit

Fluidkey drained for ≥$1.58M as logging-tool flaw bypasses local keys

Bitquery traces ≥$1.58M across six chains to Fluidkey attacker-linked wallets on Nov. 19, after a logging-tool flaw let the relayer trigger transfers when users opened the app.

by 4 min read

Fluidkey, a stealth-address privacy wallet, was drained in a ~36-minute window on November 19 despite private keys never leaving users' devices. Bitquery has traced at least $1.58 million out of user accounts across six chains between 17:28 and 18:04 UTC, with the attacker funneling stolen assets to collector wallets and Tornado Cash. Sources: The Defiant and the expanded tracing writeup by Spendnode / CryptoSlate.

What happened

Fluidkey gives each incoming payment a new receiving address derived client-side, in the stealth-address pattern popularized by ERC-5564. Transactions are submitted on behalf of the user by Fluidkey's own relayer, which pays network fees. Keys never leave the user's device, but the relayer is still in the trust path for when and which transfers get signed and broadcast.

On Nov. 19, ~17:28 UTC, unauthorized outbound transfers began. Fluidkey has attributed the incident to a logging-tool flaw that let an attacker trigger transfers when users opened the app. A full technical post-mortem linking the logging-tool vector to the observed on-chain activity had not been published at the time of the trace, per Spendnode's note on the source gap.

The service suspended web, iOS and Android access during the window and advised users to avoid the app. Fluidkey restored the apps on Nov. 20 behind a required upgrade to version 1.79.22, per CryptoSlate.

Bitquery's tracing method

Bitquery counted a payout as attacker-linked when the recipient forwarded funds to a collector wallet, forwarded them to Tornado Cash, or received fee funding from the attacker. That filter gives a lower bound:

  • At least $1.58M confirmed attacker-linked across six chains.
  • Upper bound ~$1.64M per Bitquery's broader classification.
  • An additional ~$65,000 went to unclassified destinations, excluded from the lower bound.
  • Roughly $277,000 went to an older wallet in a pattern Bitquery read as one user withdrawing their own funds, and was excluded from the theft total.
  • Independent drain trackers put the figure near $1.63M across seven chains when counting suspected destinations and excluding likely owner withdrawals.

Bitquery's classifications rest on wallet histories and exchange-address labels — attribution is labeler-dependent, as always for live-tracking of fresh clusters.

Why local keys weren't enough

The post-mortem matters because it isolates a class of risk that local-key privacy wallets inherit from their relayer. In a standard EOA model, an attacker needs the key. In a stealth-address + relayer model, the attacker needs to make the relayer act on a user session — the key stays put, the signed transaction still goes out. Fluidkey's own notice points at logging-tool scope rather than a key-derivation or signature-bypass flaw, consistent with that class of attack.

Readers tracking the architectural detail should watch for Fluidkey's incident report to disclose:

  1. Which logging tool, scoped to which services (frontend telemetry, relayer logs, or both).
  2. Whether the compromise exposed session tokens that authorized the relayer to submit transactions on a user's behalf.
  3. Whether the stealth-address derivation itself was ever at risk, or whether every drain used a valid user-side signature produced during an authenticated session.
  4. The full list of attacker addresses and Tornado Cash deposit transactions; Bitquery has the cluster, Fluidkey has not published it in-line yet.

What to watch

  1. Fluidkey's technical post-mortem, scoped to the logging-tool vector and the relayer trust model.
  2. Tornado Cash deposit cadence from the attacker cluster — six-chain exits usually converge on Ethereum mixing within days.
  3. Refunds: Fluidkey has not yet committed to a reimbursement plan at the time of writing.
  4. Downstream wallets running the same stealth-address + relayer pattern (Umbra, Labyrinth, Railgun's relayer tier): any shared dependency surfaces fast after this kind of disclosure.

Context

Fluidkey marketed local key custody as a feature. The hack doesn't invalidate that claim — on the facts available, keys do appear to have stayed on-device — but it reframes what "local keys" protect you from. A relayer that submits on your behalf is a signing-adjacent dependency: it doesn't hold your key, but it can hold a session that acts like one. For the stealth-address stack, this is the first live demonstration that the relayer is in scope for user-fund risk. The pattern echoes frontend / infrastructure compromises (Curve's DNS in 2022, Balancer's frontend in 2023, the recent Fluidkey logging-tool case) rather than any protocol-layer failure.

Related stories