DEV Community

Mirage Cloud IA
Mirage Cloud IA

Posted on

Deepfake Fraud: How Small Businesses Protect Their Payments

For years, payment fraud against small businesses has relied on a convincing email: a supplier changing bank details, a director asking for an urgent transfer. What has changed is that the email can now be backed up by a phone call in the director's own voice, or a video call with faces that look right. The best-known case was reported by Hong Kong police in February 2024. An employee at the engineering firm Arup made fifteen transfers worth about 25 million US dollars after a video conference in which the colleagues on screen turned out to be fakes.

Why detection is the wrong defence
Advice to spot a flat voice or an odd blink does not hold up. The tools keep improving, a rushed employee under pressure will not scrutinise a face, and the fraud does not depend on quality. It depends on urgency and authority. The defence that works is a process that does not trust the channel a request arrives on.

Five controls that do the work

  1. Call back on a number you already had. For any payment request or change of bank details, hang up and call the person on a number from your own records, never the number the caller or the message gives you. A callback to the real person defeats cloned voices, hijacked email and fake video calls at once.
  2. Two people for large or unusual payments. Set a threshold. Any transfer above it, and any new payee, needs a second person's approval, whatever reason is given.
  3. Never change payment details because of a message. Changes go through a documented form and are verified with the supplier by phone, independently of the request.
  4. Let urgency trigger more checks, not fewer. Make it normal to say I will call you back. Attackers rely on pressure, and a policy that removes it removes their main tool.
  5. A shared code or question. For key people, an agreed code word or a question only they could answer can confirm an identity on an unexpected call. Keep it offline, and change it after use.

Make it part of normal life
➟ Put the process on one page and give it to everyone who can move money, including temporary staff and freelancers.
➟ Run a drill: send a test request and see whether staff follow the process, without blame.
➟ Limit how much public audio and video of key people is available. A short sample of a voice is enough to clone it.
➟ Be careful with out-of-office messages and posts that reveal travel plans, because attackers time their requests around them.
➟ Ask your bank about its controls: daily limits, alerts and dual authorisation. Euro transfers at most banks now include a check that the payee name matches the account, so never override a mismatch warning without verifying.

If it happens
➟ Call your bank immediately and ask for the transfer to be recalled. The first hours matter most.
➟ Report it to the police, and use the national platform for victims of cyber incidents, cybermalveillance.gouv.fr, for guidance.
➟ Keep the evidence: emails, call logs and any recordings.
➟ Tell your insurer, and check whether your policy has cyber or fraud cover and a notification deadline.
➟ If personal data was exposed, remember the 72 hour GDPR notification rule.
➟ Review the process afterwards and fix the gap that was used.

Where AI helps
On the defensive side, AI can flag unusual payment requests, summarise supplier changes for review and help you draft the one-page process. Mirage Cloud connects to Qonto and includes a finance agent scoped to cash position and budgets. None of that replaces the callback and the two-person rule, which have to stay human.

Top comments (0)