How to Spot AI Scams, Deepfakes and Voice Clones

Do not rely on spotting flaws in a picture or a voice, because those flaws fade as the tools improve. Check the situation instead: pause when a message is urgent or secret, and confirm it through a second channel you already trust, such as calling back on a number you already have.

Can I tell a fake image, voice or video by looking or listening?

Not dependably. Advice such as "count the fingers" or "listen for a robotic tone" described real flaws in some fakes once and is already out of date for many, and a scammer needs only one attempt that works.

Real photos can look odd from compression, so a flaw proves nothing, and neither does a flawless clip. Check instead who is asking, what they want and how they push you to act.

What do AI scams look like now?

AI has mostly made old scams cheaper, faster and more convincing, not invented new ones.

• A cloned voice in trouble. A call from "your son" or "your grandchild" reports an accident or an arrest and wants money now. A short public clip can be enough to imitate a voice.
• A boss who needs a payment. A message, or even a video call, seems to come from a manager asking for an urgent transfer, gift cards or new bank details.
• Polished phishing. Bad spelling used to give scam emails away. Now they can be fluent and use real details from your public profiles.
• A famous face selling an investment. A video of a well-known person praising a scheme. Check their official channels; never invest through an ad's link.
• Fake profiles and reviews. A friendly stranger with a believable photo soon asks for money or a private chat, and praise can be generated in bulk, so glowing reviews prove little.

How do I check who is really contacting me?

Slow down first. Urgency and secrecy are the scammer's tools, and "do not tell your parents" or "your account will be frozen today" is pressure, not evidence. Then verify through a channel you already trust:

• Hang up and call back on a number you already have, not the one that rang or the one in the message, since caller ID can be faked. For a "bank", use the number on your card. On a landline, use another phone or wait a few minutes, as a scammer can sometimes keep the line open.
• Agree a code word with your family. Choose something never posted online and share it in person; a personal question alone is weak, since the answer may be on social media. A caller who cannot give it is unverified, not proven fake, so call the person back on their own number.
• If you cannot reach them, try someone who is with them, or the police. Do not take the caller's word that they are in danger.
• Use a second person for money. Confirm new bank details or an unusual transfer with the supposed sender on a number you already hold, or with another colleague, however senior the requester seems.

Do not pay, share a one-time code or install software because of an unexpected message. Gift cards, cryptocurrency and rushed transfers to a new account are especially hard to reverse. A real friend, boss or bank has no reason to refuse "I will call you straight back".

How do I check whether an image or claim is real?

Ask where it came from, not whether it looks right.

• Find where it first appeared. A reverse image search, built into many browsers and photo apps, can show earlier uses of a picture, such as a "breaking" photo that is really years old. No earlier results does not make it real; it may be new.
• Find the original source. A screenshot of a post is not the post. Find the account or article itself, note its date and whether the account is new or anonymous, and see whether established outlets confirm it independently.
• Check the details. Do the landmarks, signs, language, weather and date fit the place and day claimed?

How to fact-check an AI answer shows how to confirm a claim against a primary source.

Can an AI detector tell me whether something is fake?

A detection tool can be one clue, and no more. These tools err in both directions, calling a real photo fake and passing a fake as genuine, and results shift after a file is cropped or re-saved. Some tools add a label or hidden mark to AI-made images, but it can be lost when a file is screenshotted or edited, so a missing label proves nothing.

An assistant can help you notice pressure tactics in a suspicious message, but it cannot see who really sent it. Paste only the text, never a one-time code or a password; the AI assistant privacy checklist covers what to hold back.

A "looks real" result is no reason to send money, and a "probably fake" one is no reason to accuse a classmate or colleague. Use a verdict only to decide what to check next.

What should I do if I think I have been scammed?

Act quickly, and do not be embarrassed. These scams work on careful people.

• Contact your bank or card provider at once, using the number on your card, to stop or recall the payment and replace a card whose details you shared. Whether money can come back depends on the payment type, how fast you tell them and where you live.
• Change passwords, starting with your email and any account that shares the password, and turn on two-step sign-in. If you gave remote access or installed software, disconnect the device from the internet and get it cleaned.
• Keep the evidence (messages, screenshots, numbers, payment details) and report it to the platform and to the police or consumer-protection body where you live.
• Expect a second wave. Anyone who later offers to recover your money for a fee is most likely running another scam.

This is general guidance, not legal or financial advice.

How can I help older relatives and children?

Talk before anything happens, and without blame, because a person who fears being scolded hides a scam.

For an older relative, agree the code word and one rule: if anyone asks for money urgently, even if it sounds like me, hang up and call me back. Keep the bank's real number on a card by the phone and name one person to ask before paying anything.

Children meet fake profiles, hijacked friend accounts and fake images of classmates. Teach them that a voice or photo can be faked and that they can tell an adult without being in trouble; if someone threatens to share a picture unless they pay, they should not pay and should keep the evidence. AI chatbots and kids has wording for the no-blame promise.

Where NAMELESS stands

NAMELESS is an AI model still in training and not yet available, with no launch date. Our goals for it are unlimited context, no rereading of the conversation and stronger reasoning, but none of that would let an assistant confirm who is on the other end of a call; that still takes a second channel and your own judgment. To follow progress, join the waitlist.