Deepfakes and scams

You get a voice message from your cousin. It's really their voice. They ask for money, fast. Is it them? Have a guess.

The whole story in three sentences. A deepfake copies someone's voice or face with AI. It's hard to tell from the real thing. Scammers use it to make old tricks more convincing.

Think of a costume party with perfect masks. You can stare at a mask all night, and still not know who's behind it. So you ask something only the real person knows.

• A copied voice. In "help request fraud", a "son" or "daughter" asks for money by app. Scammers can copy their voice with AI.
• A fake video. Fake ads use famous people, sometimes in deepfake videos, to sell investments.
• A fake picture. AI makes realistic images in seconds. Not every one is labelled.

The tricks are old: haste, money and secrecy. AI only makes the mask better.

A voice model predicts sounds, an image model predicts pixels. That's the rule.

Give it enough examples of a voice, and it predicts how that voice sounds. Saying words the person never said. The result is a prediction too, but a very good one.

That's why looking for mistakes isn't enough. The Fraudehelpdesk writes that a deepfake is hard to tell from the real thing.

There are rules. In the EU, deepfakes must be labelled. And some makers add an invisible watermark. Google does that with all images its tool makes. But scammers don't follow the rules.

Four messages. Real or fake? And what do you do?

Everything runs in your browser. Nothing you type is sent anywhere.

Four steps. The first one matters most.

Notice the pressure

Haste, money, and "don't call me". The Fraudehelpdesk names these signs. Promised high returns are a sign too.

Stop before you pay or click

Never pay after a message alone. Don't click the link in an ad. A real person can wait five minutes.

Check through a channel you know

Call the person on the number in your own contacts. Not the new number from the message. Ask if they really sent it.

Agree a code word

Agree a code word or a special question with your family. Use it whenever someone asks for money.

Three people meet a mask.

• Don't rely on spotting mistakesFakes are hard to tell apart.Check through a channel you know.
• Don't call back the number in the messageThat's the scammer.Use the number in your own contacts.
• Don't share a shocking video right awayIt may be fake.Check first, or don't share.

• organisationFraudehelpdesk. Fraudevormen en -termen.A deepfake copies someone's voice or face with AI and is hard to tell from the real thing. In 'help request fraud' a 'son' or 'daughter' asks for money by app, sometimes with a recorded copy of their voice.
• organisationFraudehelpdesk. Hulpvraagfraude.Scammers pose as a family member on WhatsApp, from a 'new number', and ask for money for an urgent problem. Signs: haste, a promise to pay back fast, and no call or visit allowed. Scammers can copy a voice with AI. Tips: never pay after a message; call the person on the number in your own contacts; agree a code word with family.
• organisationFraudehelpdesk. Ik heb geklikt op een nepadvertentie met een bekende Nederlander.Fake ads use famous Dutch people, sometimes in deepfake videos, to sell investments. Promised high returns are a sign it's not true. (Checked through a search engine.)
• governmentEuropean Commission. AI Act.The EU AI Act sorts AI into four risk levels. Some uses are banned, like social scoring. People must be told when they talk to a chatbot, and deepfakes must be labelled. AI literacy rules apply from 2 February 2025.
• documentationGoogle AI for Developers. Image generation with Gemini.Describe a scene in rich detail: the more specific, the more control. Be specific about the style. Improve an image in a conversation. You can add, remove or change parts, or change the style. All images it makes carry a SynthID watermark.

All sources of this course →

You get 7 questions about the whole page. Get 5 right, and you've mastered the topic.