Skip to content

Knowledge base

Spotting AI fraud and deepfakes at work

By Bob Biesemans · published on

Flawless phishing emails, a cloned voice of your boss, a video call full of deepfakes. Which warning signs remain, and which agreements protect your team.

Short answer

AI makes phishing emails flawless and can imitate voices and faces, so spelling mistakes are no longer a sign. Look at the request itself: urgency, secrecy, an exception to procedure or a new account number. Verify through a channel you choose, never change bank details based on one message, and have large payments approved by two people.

What AI changes about fraud

For a long time you could spot a phishing email by its spelling mistakes, an odd greeting or a sloppy logo. With AI, fraudsters write flawless emails in any language, in the tone of your bank, your supplier or your own management. They can also imitate a voice or create video of someone you know.

The Centre for Cybersecurity Belgium warns that someone without technical knowledge can set up a phishing operation within days using AI. In March 2026 Safeonweb received more than 42,000 reports a day on average. Anyone still looking for spelling mistakes misses exactly the most dangerous messages.

What it costs

According to Febelfin, people in Belgium lost about 93 million euros to phishing in 2025, across more than 11,000 cases. Febelfin explicitly singles out targeted attacks using AI and cloned voices.

Companies are just as much a target. At the engineering firm Arup, an employee paid about 25 million dollars after a video call in which the finance director and colleagues turned out to be deepfakes. The fraud worked because everything looked real and the request came from someone with authority.

The warning signs that remain

Look at the request, not the form. Urgency, secrecy, an exception to the usual procedure, a new account number or a different channel than normal: those are the signs AI doesn't polish away. A familiar voice or face is no longer proof that the person is real.

Watch out for AI in your own work too. An AI summary of an email repeats what the email says, including a request to change an account number. Decide on payments and changes based on the original and a check, never on a summary.

Agreements that protect your team

Verify every unusual request through a channel you choose yourself: call back on the number you already had, not the one in the email or call. Never change a supplier's or client's bank details based on a single message. Have payments above an agreed amount approved by two people.

Make it easy to report, and to say 'no' to a manager in a hurry. Anyone who doubts and checks is doing the right thing, even if the message turned out to be genuine. In Belgium, forward suspicious messages to suspect@safeonweb.be.

Measuring whether your team recognises it

Recognising AI fraud doesn't depend on age or job title but on what someone does in the situation. You measure it with realistic cases: a flawless email with an urgent payment request, a call with a familiar voice, a changed account number in an AI summary. Measure everyone the same way, explain where needed and measure again. Record what you did as a measure for your staff's AI literacy under Article 4 of the AI Act.

Source: Febelfin, actieplan voor de aanpak van online fraude (juli 2026), VRT NWS, actieplan van de banken tegen phishing (8 juli 2026), VRT NWS, phishingmeldingen bij Safeonweb (7 april 2026), Dezeen, Arup verliest 25 miljoen dollar na deepfake-videogesprek (17 mei 2024)

Frequently asked questions

How do you spot a phishing email written with AI?

Not by the language any more. Look at what the email asks: urgency, secrecy, a new account number, logging in via a link or an exception to the usual procedure. If in doubt, check through a channel you choose yourself.

What should you do if your boss calls with an urgent payment request?

A familiar voice can be cloned too. Hang up and call back on the number you already had, then follow the usual approval procedure. A real manager understands that check.

Where do you report a suspicious message in Belgium?

Forward suspicious emails to suspect@safeonweb.be and report it internally so colleagues are warned. If a payment has already been made, contact your bank straight away.

Further reading

Read next

Help us improve the website?

With your permission, Google Analytics, Microsoft Clarity and Vercel Web Analytics measure visits to our public pages. Your test answers, results and dashboard are not sent. We do not use advertising cookies. Privacy policy