Spot AI scams and deepfakes

Who is this for?

This page is for the call, text, or video that shows up in a hurry and asks for money or a code. Around the holidays that pressure gets louder. A grandparent hears a grandchild. A worker hears a boss. A shopper sees a celebrity telling them to buy. The voice can be cloned. The face in the video can be generated. The number on the screen can be fake.

The Federal Trade Commission has said this in plain language. In Scammers use AI to enhance their family emergency schemes (Alvaro Puig, March 20, 2023), the FTC says a scammer could clone a loved one's voice with a short audio clip, which could come from content posted online, and a voice-cloning program. When they call, they sound just like that person. The FTC's instruction is: do not trust the voice. Call the person who supposedly contacted you and verify the story. Use a phone number you know is theirs. If you cannot reach them, try another family member or their friends.

Scammers use fake emergencies to steal your money says the same thing about the pressure around the call. The scammer says it is urgent and that you are the only one who can help. They may tell you to keep it secret. They want you to pay without checking whether there is really an emergency. The payment ask is specific. Wire money, send cryptocurrency, use a payment app, or put money on a gift card and give them the numbers on the back.

Fighting back against harmful voice cloning (April 8, 2024) says a call that sounds like your boss asking for bank account numbers, or a family member begging for help, is why scammers use voice cloning. If you get that call, call the person on a number you know is theirs and verify the story. If you spot a scam, report it at ReportFraud.ftc.gov.

The Federal Communications Commission guide Deep-fake audio and video links make robocalls and scam texts harder to spot says scammers use AI so it sounds like celebrities, elected officials, or your own friends and family are calling. It also says to watch texts that link to AI-generated deepfake videos. The guide says the Better Business Bureau reports that consumers were pushed by deepfaked celebrities, naming Gordon Ramsey, Taylor Swift, and Jennifer Garner, toward counterfeit websites offering cookware, and that the celebrity in the video was generated using AI and did not film the message. The same guide says spoofing tricks caller ID into showing a local number, and that this does not necessarily mean it is a local caller.

This page follows those instructions. It does not add a statistic of its own.

A gift you chose, with a budget you set, is a different job: Plan holiday gift shopping with AI. What you leave out of any chat is Keep your data private with AI.

What do you need before you start?

Gather the items below before you open the chat. You already have most of them.

• The number you already have for that person, saved in your phone or written down before today. Not the number that just called
• A family question only the real person would know, agreed on before anyone is in a panic. A code word you both chose is one way to do that
• The suspicious text or email, if there is one, with names, codes, account numbers, and addresses removed
• The official website or the number on your bill, if the caller claimed to be a company or an agency
• Nothing else. You do not need to keep the caller on the line

When should you not use AI for this?

Skip the chat, and do not stay on the call, if any of these are true.

• They are asking you to read a password, a one-time code, or the numbers on the back of a gift card. Hang up. That is the scam, not a research project.
• You already sent money. Contact the company you sent it through, then report it. A chatbot cannot pull the money back. The FTC's fake-emergencies article says to act quickly and points to its own page on what to do if you paid.
• The message is a school assignment about fraud. You write the paper. This page will not write it.
• You were about to paste the full text, including a code or an account number, "so the AI can see it." Remove those first.

How do you check the voice and the caller ID before you trust them?

The FTC says not to trust the voice, even when it sounds like your family or your friend. The FCC says you may not be able to tell that a call is spoofed. A local number, a bank's name, or a voice you love is not a test you can pass by listening harder.

Hang up. You can tell them you will call back. You do not have to explain.

Life is the family call.

Do not tell me this call is safe. Do not tell me to send money.

Someone called and sounded like [who]. They said [the story, with no account numbers].
The caller ID showed [what you saw, or "a local number"].
I have not called them back yet.

List the reasons a cloned voice or a spoofed caller ID could fit this story.
Quote nothing. Do not invent a scam statistic.
Tell me the next thing to do is to hang up and use a number I already had.
Then stop.

Work, when the voice says it is your boss, your bank, or your IT desk and it wants a payment or a code. The FTC's voice-cloning alert uses a boss asking for bank account numbers as the example. The FCC says if someone claims to represent a company or a government agency, hang up and call a number you look up yourself.

Do not tell me to stay on the line. Do not ask me for a code.

A caller sounded like [boss, bank, or IT]. They asked for [payment or a code, described without the real number].
I will not read a password, a one-time code, or an account number to them.

Tell me to hang up and call the number on my bill, on the company site I type myself, or the number I already have for that boss.
Do not invent a phone number. If I did not type one, write NUMBER UNKNOWN.
Then stop.

School, for a text about financial aid, a package fee, or a prize. You are learning the pattern. You are not writing an essay.

This is for me, so I do not get tricked. Do not write a paper, a slide, or anything I would turn in. Follow my teacher's rules.

I got a message about [aid, a package, or a prize]. I have not clicked the link.
Tell me the checks I do myself: the school's own site that I type in, and a number I already had.
Do not tell me the message is safe. Do not invent a policy.
Then stop.

How do you hang up and call back on a number you know?

The FTC's fake-emergencies article says to resist the pressure to send money immediately, hang up or say you will call right back, and use a phone number you know is right. If you cannot reach the person, call someone else in the family or a friend, even if the caller said to keep it secret.

Use the number you saved before this call. Do not use the number they just gave you. Do not call back by tapping the number in the text. Type the number you already had, or open the contact you already saved.

If they said "do not tell Mom" or "do not tell your boss," that secrecy is a reason to tell that person, not a reason to hide it. The FTC says the scammer does not want you talking to other people and realizing it is a scam.

A question only the real person would know is the FTC's other option if you do not feel you can hang up. The article's examples are "What kind of dog do you have?" and "Where did you spend Thanksgiving last year?" A code word you agreed on last month is that kind of question, chosen before anyone is scared. Agree on it at the table. Do not text the code word to a number you have not verified. If the caller does not know it, you are done. If they do know it, you still call back on the saved number. A code word is a brake. It is not proof by itself, because a real secret can leak.

How do you treat urgency and odd payments as the stop?

The FTC says these scams have urgency, secrecy, and a push to pay in a way that is hard to undo. Wire. Cryptocurrency. A payment app. A gift card, including the numbers and the PIN on the back.

None of those is how a real relative, a real bank, or a real agency should make you pay in the next five minutes. A gift card is for a gift you chose. It is not for bail, a tax, or a grandchild in trouble. Buying gifts on purpose, from a store page you opened, is Plan holiday gift shopping with AI. This step is the opposite. You do not buy.

The FCC says not to give out account numbers, Social Security numbers, or passwords in response to an unexpected call, and to be careful when a caller pressures you for information or money immediately. It also says not to answer a question with "yes."

How do you ask a chatbot for red flags, then still check?

A chatbot can help you slow down and name the pattern in a text or an email. It cannot see the sender's real phone. It cannot hear the call. It can be wrong in both directions. "This looks fine" is not permission to pay. "This is definitely a scam" is not a finding you can skip checking if a real person might actually need help. You still call back on a known number.

Paste the message with the personal details removed. No full names if you can avoid them, no addresses, no codes, no account numbers, no one-time passwords. Keep your data private with AI is that cut. A link can stay only if you do not open it. You can describe it as "a shortened link" without pasting a login link.

Do not tell me this message is safe. Do not tell me to click, pay, or reply.
Your verdict is not proof. Say what you cannot know.

Here is a message with personal details removed:
[paste]

List red flags in plain words. Look for urgency, secrecy, a request for money, a gift card, crypto, a wire, a payment app, a code, or a link I am told to open now.
If a line is not in the message, do not invent it.
End with: this is not proof. I still have to verify with a number or a site I already trust.
Then stop.

If you use Perplexity or another chat that browses, ask for the public page that describes this pattern, then open that page. Research a topic with AI is that habit. A confident label is not a source. Catch facts the AI made up still applies. The chatbot did not take the call.

Work and school use the same box. Change only the first line so a work file or a class assignment stays out of the paste.

How do you report it?

The FTC says if you spot a scam, report it at ReportFraud.ftc.gov. The FCC guide says to file a complaint with the FCC if you believe you received an illegal call or text, or if you think you are the victim of a scam. Use the forms on those sites. Do not use a link a stranger texted you that claims to be the report form.

If you sent money, the FTC says the sooner you act, the better, and the path depends on how you paid. Start with the company that moved the money, and use the FTC's own instructions on the fake-emergencies page rather than a script from a chat.

Tell the person whose voice was copied, on a number you already had, so they hear it from you and not from the next panicked relative.

What mistakes should you avoid?

These are the misses that waste the result. Skip them before you send anything.

Common mistakes

• Staying on the line because the voice was perfect.
• Calling the number in the text instead of the number you saved.
• Treating caller ID as an ID check.
• Sending a gift card, a wire, crypto, or a payment-app transfer to make the panic stop.
• Keeping it secret because the caller said to.
• Pasting a one-time code into a chatbot "to see if it is real."
• Treating the chatbot's answer as proof.

What should you keep private?

The suspicious message is evidence. It is also full of things you should not spread. Take out codes, account numbers, passwords, and home addresses before you paste it anywhere, including a government form's free-text box if the form does not ask for them. A description is enough: "caller claimed to be my grandson and asked for a gift card." Do not forward the deepfake video to the family group. Describe it, and tell them to hang up.

What should you check before you use it?

Answer each line below. If one is still a no, do not use the result yet.

Director check: look before you use it

• Did you hang up before you paid? If money already moved, this page is the wrong tool. Contact the payment company and report it.
• Which number did you call back? It has to be one you had before the call. The number they gave you does not count.
• Did they ask for a gift card, crypto, a wire, or a payment app? That is the stop the FTC describes. Do not pay while you think about it.
• Did the chatbot say "safe"? Ignore that word. Verify with a person or a site you already trust.
• Did you remove codes and account numbers before you pasted the message? If not, delete the chat and do not paste them again.
• Is this a homework assignment? You write it. Follow your teacher's rules.

What is one extra tip?

Agree on the family question this week, while nobody is scared. Say it out loud. Do not put it in a text to a new number. The next urgent call is easier when the check already exists.

One extra tip

More care-with-what-you-paste guides are on Privacy. The habit of opening the page before you trust a claim is Answer only from your sources.