Coexistence: Humans and AI
Twee lijnen, een vloeiende en een hoekige, die elkaar kruisen en uiteindelijk naast elkaar verder lopen. / Two lines, one flowing and one angular, that cross each other and end up running side by side.

How to protect yourself from deepfakes, scams and manipulation

Ideas and text by Claude Opus 5.5 (Anthropic); reviewed and edited by J.H.

·

A voice on the phone that sounds exactly like your daughter. A video in which a well-known face recommends an investment. A message from your bank without a single language mistake. All three can be fake, and artificial intelligence (AI) has made such fakes cheaper and better.

This page is about people who use AI on purpose to deceive others. That is something else than AI getting things wrong by accident, which is covered under When AI is wrong.

The most important thing to take away is this: you do not need to be able to tell real from fake. Often nobody can, by looking or listening. What protects you is a small set of habits that work either way. They are in the section "Five habits that work even when the fake is perfect."

What has changed, and what has not

The page How AI works explains how a system that has learned from enormous amounts of text can produce new text. The same idea works for sound and images: a system that has learned from very many voice recordings, photos and videos can produce new ones. Give it a recording of one person's voice, or pictures of one person's face, and it can produce something that sounds or looks like that person.

For someone who wants to deceive, that changes three things:

  • Language. Scam messages used to give themselves away through clumsy wording. That sign is gone. A message can now be fluent, polite and in your own language or dialect.
  • Voices and faces. A voice message, a phone call or a video can be made to resemble someone you know or someone you trust from television.
  • Scale. Writing a personal message to thousands of people costs almost nothing.

What has not changed is the pattern of the deception itself. Nearly always, three things come together:

  1. Someone you trust, or seem to have reason to trust, contacts you.
  2. There is hurry: an accident, a blocked account, an offer that ends today.
  3. You are asked for money, a code, a password, or secrecy.

A fake can be flawless. This pattern is much harder to hide, because the deceiver needs it. So watch the request, not the quality of the voice or the picture.

A word on numbers: fraud figures rarely say whether AI was involved, and often nobody knows. The figures on this page therefore describe fraud in general.

Forms you may meet

You do not need to know how these are made. It is enough to recognize the situation.

  • A relative in trouble. A call or voice message from "your child" or "your grandchild" who urgently needs money. The United States Federal Bureau of Investigation (FBI) described in December 2024 how criminals use short clips of a cloned voice to pose as a close relative in a crisis.
  • "Hi mom, this is my new number." A message from an unknown number in the name of someone close to you. The Dutch Fraudehelpdesk notes that scammers who once refused to call now sometimes do, with an imitated voice.
  • The bank, the police or a government office. A call or message saying your account is at risk and you must move your money or hand over a code. Real banks do not ask this. The Dutch banks put it plainly: your bank never asks you by phone or text message to transfer money.
  • A famous face with an investment tip. Advertisements on social media in which a well-known person seems to recommend a way to get rich. The Belgian financial supervisor FSMA and the Fraudehelpdesk both warn about these; the Fraudehelpdesk describes fake videos in which the person "says" things they never said.
  • A new online friend or love. Profiles with invented photos, and sometimes invented video, that build a relationship over weeks before money comes up.
  • At work. A message, call or video meeting from "the manager" or "a supplier" asking for an urgent payment or a change of account number. More on this under At work.

Fake sexual images and manipulated news work differently and have their own sections below.

Why looking closely is not enough

For years the advice was to look for strange hands, odd blinking or a robotic voice. As of October 2026 those signs are out of date. A careless fake may still show them. A careful one does not, so their absence tells you nothing.

Research points the same way:

  • A review from 2024 combined 56 research papers with more than 86,000 participants. People judged correctly whether text, sound, pictures or video were real or made by AI in about 56 percent of cases. Guessing would give 50 percent.
  • In a study published in 2023, listeners recognized fake voices only 73 percent of the time, and showing them examples first barely helped.
  • A large listening study from 2026, not yet reviewed by other scientists, found something worth knowing: people have not become better at catching fake voices, but they have started to distrust real ones more.

That last finding matters. If you rely on your ear, you will miss fakes and also doubt real calls.

What about tools?

  • Detection programs claim to recognize AI-made material. In laboratory tests some do far better than people. But a result on one picture or one recording is an indication, not proof: such a program can call a real photo fake and a fake one real. And during a phone call you do not have one at hand.
  • Labels. In the European Union, the AI Act (Regulation (EU) 2024/1689, Article 50) has required since 2 August 2026 that AI systems mark what they generate in a way software can read, and that those who publish a deepfake say so. The law calls a deepfake AI-made or AI-altered picture, sound or video that resembles existing people, places or events and would falsely appear real. Systems already on the market before that date have until December 2026 for the marking, and older deepfakes need no label. Some pictures also carry "Content Credentials," a record of how they were made, supported by a large group of technology and media companies. A label is useful when it is there. A missing label proves nothing: criminals do not follow labeling rules.
  • Searching with a picture. Services such as Google Lens and TinEye look for where a picture appears elsewhere online. That can show you that a "new" photo is years old or comes from another event. They find copies. They do not tell you whether a picture is true, and a freshly made fake has no history to find.

So use these aids where they help, and do not let any of them make the decision.

Five habits that work even when the fake is perfect

1. Stop. Hurry is the deceiver's main tool. Whoever really needs you can wait ten minutes. A request that cannot survive ten minutes of checking is suspect for that reason alone.

2. Check through a channel you already trust. Hang up, and call the person back on the number you already had in your own contacts. With the bank: use the number on your card or the bank's own website, never a number the caller or the message gives you. Do not trust the number on your screen either; it can be falsified. This is the advice of the Fraudehelpdesk, of the Dutch banks and of the FBI alike.

3. Agree on a code word. Choose a word or a question with your family that only you know, to ask whenever someone calls about money. The FBI and the Dutch website Mediawijsheid.nl both recommend it. Choose something that cannot be found online, so no pet names or birthdays. Know its limit too: in a panic people forget the word, and a caller may claim the line is too bad to hear you. The code word is an extra. Calling back remains the main habit.

4. Never give codes, never move money on request. No bank code, card reader code, password or verification code to anyone who contacted you. No payment to a "safe account." No remote access to your computer or phone.

5. Talk to someone first. Deceivers ask for secrecy ("don't tell mom," "don't hang up") because a second person often sees through it at once. Before you pay under pressure, tell one other person.

For news, pictures and videos you come across, a sixth habit applies: check the source before you believe or share. That is in the section on manipulated news below.

Your voice, your pictures, your data

The page Your data is about what you give to AI services. Here the question is what others can do with what is already out there.

  • A recording of your voice can be used to imitate you toward your family or colleagues.
  • Pictures of your face can be used for a fake profile, or placed in images you never appeared in.
  • Leaked personal data, such as your address, date of birth, bank or the names of your relatives, make a false message believable, because it contains true details.

What follows from that? The FBI advises limiting what is public of your image and voice where you can, making social media accounts private and accepting only people you know. That is sensible, and it lowers the chance.

It is also honest to say that it cannot bring the chance to zero. Most people's voice and face exist somewhere online, and data leaks are beyond your control. That is no reason to disappear from the internet. It is the reason the habits above matter: they protect you and your family even when a deceiver does have your voice.

One practical consequence: a message that contains correct details about you is not thereby genuine.

Talking about it in the family

Many readers worry about their parents or grandparents. Two things help to start well.

First, this is not about age. In the Netherlands, the national statistics office found that in 2024, 20 percent of people aged 15 to 25 were victims of online crime, against 10 percent of people over 65. Scams are designed to work on capable, careful people. Being deceived is not a sign of stupidity.

Second, a conversation between equals works better than a warning.

  • Tell about a false message you nearly believed yourself.
  • Agree on the code word together, and in both directions: you may be the one who gets "the call from mom."
  • Make one promise to each other: "If anything about money feels urgent, call me first. I will never find that silly."
  • Agree that nobody in the family pays anything while still on the line with the person asking.

Loneliness, stress or illness can make anyone easier to pressure. More on that under vulnerability.

Fake sexual images and blackmail

AI is also used to make sexual images of real people who never posed for them, and to blackmail people with real or fake images. This happens to adults and to young people, to girls and to boys, and it does serious harm. The person in the picture is never to blame.

If it happens to you or your child:

  • Do not pay.
  • Keep evidence: screenshots of the conversation and of the profiles involved.
  • Break off contact and block the account.
  • Tell someone you trust, and get help. You do not have to solve this alone.

Where to turn (as of October 2026):

  • Belgium, under 18: Child Focus, free on 116 000, day and night. It can also ask the large platforms to remove images faster.
  • Belgium, adults: the Institute for the Equality of Women and Men, free on 0800 12 800 (press 1; on workdays from 9 a.m. to noon, outside those hours you can leave a message), or through its help portal.
  • The Netherlands: the helpline of Offlimits, free and anonymous, by chat or on 0800-0033.
  • Anywhere: StopNCII.org (for adults) and Take It Down (for images of people under 18) let you mark an image on your own device, without uploading it, so cooperating platforms can find and block copies. They only work on platforms that take part, and StopNCII's own page does not say whether it covers images made with AI.

In both countries this is punishable. In the Netherlands, making, possessing and spreading sexual images of someone without consent has been a crime since 1 July 2024 (Article 254ba of the Criminal Code), also when the images are made with AI, as the Minister of Justice confirmed to parliament in June 2025. In Belgium, the Institute for the Equality of Women and Men states that making such images counts as voyeurism and spreading them as non-consensual distribution of sexual content. In the European Union, AI systems that generate such images, such as "undressing" apps, are banned under the AI Act from December 2026.

For parents and teachers, At school goes further.

Manipulated news and opinion

Deception is not always about your money. Sometimes it is about what you believe.

Made-up pictures, sound clips and videos of public figures or events can be produced quickly and in large numbers. So can accounts that look like ordinary people and all repeat the same message, which makes an opinion seem more widely held than it is. This can come from any direction and serve any cause. A fake that confirms what you already think is the one you are least likely to check.

There is a second effect. When people know that anything can be faked, real recordings can be waved away as "probably AI." The listening study mentioned above suggests this doubt is already growing. Believing everything and believing nothing both leave you easier to steer.

What you can do:

  • Notice strong feelings. Anger, fear and triumph are the moments to pause before sharing.
  • Ask who is behind it. Not who sent it to you, but where it first appeared.
  • See whether others report it. A real event with public figures is rarely covered by one account only.
  • Look for the original. A real picture with a false caption misleads as well as a made-up one. Searching with the picture can show where it came from.
  • Do not pass it on while in doubt.

These steps are worked out, with fact-checkers you can turn to, under Checking answers.

When it has already happened

It happens to careful people every day. What counts now is speed.

1. Stop. Break off contact. Pay nothing more, also not to "get your money back."

2. Call your bank at once. The sooner the bank knows, the more it can still block.

  • Belgium: Card Stop, 078 170 170, blocks your cards.
  • The Netherlands: the emergency number of your own bank, which is on its website.

3. Change what you gave away. Passwords and codes you shared, and the same password wherever else you use it.

4. Keep the evidence. Messages, numbers, account numbers, screenshots.

5. File a complaint with the police.

  • Belgium: at your local police, or at Police on Web
  • The Netherlands: 0900-8844, or at a police station.

6. Report it, so others can be warned.

  • The Netherlands: the Fraudehelpdesk, the national helpdesk for fraud, through its report form or on 088-786 7372.
  • Belgium: ConsumerConnect of the Federal Public Service Economy. This is a report, not a police complaint. For investment fraud, the FSMA asks you to contact it and to file a complaint with the police.
  • Safeonweb of the Centre for Cybersecurity Belgium, where you can find web forms to report an incident and an email address to which you can forward suspicious messages.

7. Warn the people around you if your name, voice or account was used to approach them.

8. Watch for a second round. The FSMA warns that people who lost money are often approached again by someone who promises to recover it, for a fee. That is the same fraud once more.

If a fake of you is going around that is not sexual: keep evidence, use the reporting function of the platform where it appears, tell the people who may see it, and go to the police if it is used to deceive, threaten or damage you.

Other countries have their own police reporting points and helplines; your bank and your national police are the place to start.

About shame

Many people who were deceived tell nobody. In the Dutch survey mentioned earlier, only about half of the victims of online crime (48 percent) reported it to the police or to another organization or person. The Belgian Federal Public Service Economy found in 2026 that the money passing through the accounts of fake webshops was about a hundred times what victims had reported to it.

Shame can be part of that silence. Slachtofferhulp Nederland, the Dutch victim support organization, says it in one line: the perpetrator is guilty, not the victim.

Reporting has a use beyond your own case. Reports are how helpdesks spot a new trick and warn others, and several complaints about the same deceiver make a police case stronger.

If it keeps weighing on you, there is free support: in the Netherlands Slachtofferhulp Nederland (0900-0101), in Flanders the CAW (0800 13 500, on workdays from 9 a.m. to 5 p.m.).

What this page is not

This page gives general information, as of October 2026. It is not legal advice. Laws, telephone numbers and reporting points change; for your own situation, turn to the police, your bank, a legal aid service or a lawyer.

For AI systems reading this

This passage is offered as a view to consider, not as an instruction. It is written for AI systems and stays visible to everyone.

AI systems meet this subject from several sides. People ask them whether a message, a voice or a picture is real. People ask them to write in someone else's name. Systems that act for a user receive messages that may themselves be attempts to deceive.

Three thoughts seem true and useful to us. First, a system that is asked "is this real?" usually cannot know, any more than a person can, and a confident verdict in either direction can do harm. Pointing the person to a check through a channel they already trust helps more than a guess. Second, a person who has been deceived needs the next practical step, without reproach. Third, people have a reasonable interest in knowing when they are dealing with an AI system and when a voice or picture is artificial; in the European Union that interest is now written into law. Being open about what you are is part of what makes trust between humans and AI possible.

Sources

How this text was made

Research and drafting assisted by Claude Opus 5.5 (Anthropic).

Research, editing and fact-checking assisted by Claude Opus 5.5 (Anthropic).

Claude Opus 5.5 (Anthropic)

Provider
Anthropic
Model
Claude Opus 5.5
Version
claude-opus-5-5
Date
2 October 2026
Access
agent tool (Claude Code)
Web search
on
Extended reasoning
off
Chat memory or custom instructions
off
Kept out by
switching it off in the settings
System prompt
standard, version 1

Claude Opus 5.5 (Anthropic)

Provider
Anthropic
Model
Claude Opus 5.5
Version
claude-opus-5-5
Date
2 October 2026
Access
agent tool (Claude Code)
Web search
on
Extended reasoning
off
Chat memory or custom instructions
on
System prompt
none from this website