Revolut: KYC files handed to a fake government agency

The mechanism

A fintech's compliance team regularly receives official requests: court orders, investigations, regulator inquiries. Many arrive by email. Here, the email came from the agency's real domain and passed authentication checks (SPF, DKIM, DMARC). Those checks prove a message comes from a server authorized by the domain. They do not prove that the person who wrote it is entitled to make the request: a compromised mailbox at the agency produces perfectly authenticated messages. Our SPF, DKIM and DMARC guide explains what these protocols guarantee and what they do not.

Revolut blocked the address once it noticed and alerted the agency, law enforcement, data protection authorities and financial regulators. The company says its systems and customer funds were not affected. It is its second notable breach after September 2022, when a social engineering attack affected 50,150 customers.

The three flaws

1. An authenticated channel mistaken for authorization

The process accepted a request by inbound email, with no callback to a verified contact and no dedicated portal. The channel's technical authenticity replaced verification of the person and the procedure.

2. Too much data handed over

A request about a customer's identity does not need their full transaction history. Sending the whole file multiplies the impact of every mistake, and nothing in the tooling forced a narrower answer.

3. KYC data accessible without a second approval

ID document copies and biometric selfies are the most useful data for identity theft. Extracting them should never depend on a single person. The IDScan and Florida DAVID database breaches show how much attackers value these documents.

Why it concerns you

Any product that stores personal data receives outside requests: authorities, lawyers, "the customer's HR director", "the accountant", a customer who has "lost access". Each one can be imitated, and attackers research your procedures before writing: they reuse the right vocabulary, cite real laws and set a short deadline so nobody takes the time to call back. And a breach that is small in volume can be maximal in impact if it targets executives or wealthy customers.

How to test it

1. Map the exits: list every path through which customer data can leave at a third party's request: compliance mailbox, support, manual exports, GDPR requests, requests from authorities.
2. Planned verification: for each path, write down how the requester's identity and authority are verified. "The email comes from the right domain" is not an answer.
3. End-to-end exercise: with written approval from management, send the relevant team a credible fake request from an external test address, and see how far it gets before being stopped. Stop the exercise before any real data is sent.
4. KYC access: list who can view and export ID documents and selfies, and whether a second person must approve.
5. Scope: check whether the tool can send a subset (identity only) or forces the full file.
6. Traceability: find the last disclosure to a third party in your logs: who, what, when, through which channel.

How to fix it

• Official requests accepted only through dedicated portals or confirmed by calling back a contact obtained independently, never through the details in the received email.
• Two-person approval for any transfer of ID documents, biometrics or financial history.
• Send the minimum needed, with response templates per type of request.
• Named, logged access to KYC data, with alerts on unusual lookups.
• Training for compliance and support teams on institutional-looking social engineering, with real examples and a clear rule: urgency is never a reason to skip the callback.

What changes when an AI agent handles inbound requests

More and more companies hand the triage, or even the answering, of inbound email to an AI agent with access to customer files. Faced with an official-looking email, an agent is even easier to convince than a human: it is not wary of urgency or an authoritative tone, and the email's text can contain direct instructions. If the agent can attach a document or summarize a file in its reply, the Revolut attack becomes automatable. The Agent Red Team Express tests in five days what a booby-trapped email can make your agent hand over.