How to Spot AI Scams: 6 Signs of Deepfake Videos and Cloned Voices

A loved one's voice or your boss's face can now be faked with AI. Here are the deepfake warning signs listed by the FBI and an MIT research project, and how to verify a suspicious call.
Key Takeaways
- A deepfake is an image, video or audio clip created or altered with AI so that someone appears to say or do something they never did.
- The US Federal Bureau of Investigation (FBI) warned in December 2024 that criminals use AI to imitate a relative's voice in a fake crisis and demand immediate money or a ransom.
- Common deepfake video flaws show up in facial details, blinking and lip movements, and lighting and motion; with cloned voices, listen to whether the tone and word choice sound like the real person.
- The MIT Media Lab notes that there is no single tell-tale sign of AI-manipulated media and that high-quality deepfakes are not easy to discern.
- The FBI advises hanging up and calling back on a number you look up yourself, and agreeing on a secret word or phrase with your family.
AI scams are getting more convincing. The voice on the phone may sound like your child crying for help, and the face on a video call may look like your boss or a close friend, yet both can be deepfake video or a cloned voice made with AI. "It looks like them and sounds like them" is no longer enough to know who you are talking to.
This guide explains how AI scams usually work, lists six warning signs of deepfake videos and cloned voices, and shows how to verify a suspicious call and reduce the risk of being impersonated.
What Are AI Scams, and How Do Deepfakes and Voice Cloning Work?
AI scams use generative AI to make fake videos, voices, photos or messages that win your trust. Scammers often pose as relatives, celebrities, managers or law enforcement officers and then ask for money.
A deepfake puts a person's face or voice into a video or recording with AI, making them "say" things they never said. Voice cloning means using AI to imitate a specific person's voice. In a public service announcement issued on December 3, 2024, the FBI's Internet Crime Complaint Center (IC3) described several ways criminals use these tools:
- Fake family emergencies: AI-generated short audio clips of a loved one's voice, claiming a crisis and asking for immediate financial help or a ransom.
- Live video impersonation: real-time video chats posing as company executives, law enforcement or other authority figures.
- "Proof" of a real person: AI videos or photos sent in private messages to convince victims that an online contact really exists.
- Fake IDs and fake celebrities: AI-generated driver's licenses or law enforcement and banking credentials, or images of celebrities promoting counterfeit products and fake promotional videos for investment schemes.
Taiwan has seen real cases too. In 2026, Taiwan's Criminal Investigation Bureau, part of the National Police Agency, announced it had broken up a romance scam ring that used AI voice-changing software to win victims' trust. Seized items included 67 computers and tablets, with AI deepfake software among the evidence.
Under the law, Article 339-4 of Taiwan's Criminal Code was amended on May 31, 2023, to make fraud committed "by using computer synthesis or other technological methods to create false images, sounds or electromagnetic records of another person" an aggravated offense, punishable by one to seven years in prison and a fine of up to NT$1 million. Under the Fraud Crime Hazard Prevention Act, as amended in January 2026, aggravated fraud that obtains NT$1 million or more from victims carries three to 10 years in prison, with heavier penalties for larger amounts.

What Are the 6 Signs of a Deepfake Video or Cloned Voice?
In videos, check facial details, blinking and lip movements, and lighting and motion. With voices, listen to tone and word choice. In the situation itself, watch for urgent demands for money or requests for unusual payments or personal data.
The first three signs come from the FBI announcement and the MIT Media Lab's Detect Fakes research project. The last three are scam patterns described by the FBI.
1. Facial Details Look Wrong
The MIT Media Lab notes that high-end deepfakes are almost always facial transformations. Check whether the skin on the cheeks and forehead looks too smooth or too wrinkly, whether moles look real and whether facial hair looks natural. The FBI also flags unrealistic teeth or eyes and indistinct or irregular faces.
2. Blinking and Lip Movements Don't Match
Watch whether the person blinks too little or too much, and whether their lip movements look natural. Some deepfakes are built on lip syncing, and the FBI also lists voice matching as something to check.
3. Lighting, Accessories or Motion Seem Off
Do shadows fall where you would expect? When the person moves, does the glare on their glasses change angle? The FBI also lists unrealistic accessories such as glasses or jewelry, distorted hands or feet, watermarks, lag time and unrealistic movements.

4. The Tone and Word Choice Don't Sound Like Them
Voice cloning can copy how someone sounds, but not always how they talk. The FBI advises listening closely to tone and word choice to tell a real call from a loved one apart from an AI-cloned voice. For example, if a child who always calls you "Pops" suddenly says "Father," it is worth asking another question.
5. It's an Emergency and You Must Pay Now
A typical scheme described by the FBI uses a relative's voice to claim a crisis and demand immediate financial help or a ransom. The more the caller rushes you and leaves no time to think, the more important it is to stop and check.
6. You're Asked to Pay a Stranger or Share Sensitive Information
The FBI warns never to send money, gift cards, cryptocurrency or other assets to people you do not know or have met only online or over the phone, and never to share sensitive information with them. However familiar the face on the screen, a request to transfer money to a new account or hand over a verification code calls for a check first.
| Sign | What to look for |
|---|---|
| Facial details | Skin too smooth or wrinkly, unrealistic teeth or eyes, blurry face |
| Blinking and lips | Too little or too much blinking, lips out of sync with the voice |
| Lighting and motion | Shadows in odd places, glasses glare that never changes, lag |
| Tone and word choice | Tone, nicknames or phrases that don't match the real person |
| Urgency | A sudden crisis that requires you to send money or a ransom now |
| Payment and data | Requests to pay an unfamiliar account, buy gift cards or share codes |
What Should You Do If You Get a Suspicious Video Call or Phone Call?
Don't do anything the caller asks yet. Hang up and call the person or organization back on a number you already know. If you still have doubts, call Taiwan's 165 Anti-Fraud Hotline to check.
- Hang up first: End the call or video chat so you are not deciding under pressure.
- Look up the number and call back: The FBI advises hanging up, looking up the contact details of the bank or organization yourself and calling directly. The same goes for family and friends: call back on the number already saved in your contacts, not one the caller gives you.
- Ask for the code word: If your family has agreed on one, ask the caller to say it.
- Don't share data or send money: Until you have verified the caller, do not give out ID numbers, passwords or verification codes, and do not transfer money.
- Call 165: In Taiwan, call the 165 Anti-Fraud Hotline run by the National Police Agency. If you have already sent money, contact your bank and report it to the police as soon as possible.

How Can You Lower the Risk of Being Impersonated with AI?
- Agree on a family code word: The FBI recommends creating a secret word or phrase with your family to verify identity. Keep it off social media and avoid details that are easy to find, such as birthdays or addresses.
- Share less of your face and voice: The FBI suggests limiting the images and recordings of yourself online, making social media accounts private and limiting followers to people you know, so scammers have less material to build fake identities.
- Add a second lock to key accounts: Scammers often try to trick people into handing over verification codes. Turning on 2FA (two-step verification) or switching to passkeys, which need no code to type, makes account takeover harder.
- Watch for impostors after big news: After a problem at a bank or company, scammers often pose as customer service offering refunds. For ways to protect yourself, see our guide to the Bank SinoPac transfer incident.
- Talk with older relatives: Explain that both voices and faces can be faked, and agree together on who to call to check whenever an emergency call comes in.

Can You Rely on Your Eyes to Spot a Deepfake?
Not entirely. The MIT Media Lab notes there is no single tell-tale sign of AI-manipulated media and that high-quality deepfakes are hard to discern. A flaw should put you on alert, but no flaw does not prove a video is real.
Treat the six signs above as alarms: once you notice one, stop. But AI keeps improving, and fake videos with clean images and natural voices can appear too. What truly works is checking through a channel you control, such as calling a saved number, asking for the code word or meeting in person.
When a message asks for money, pause, verify and only then decide. One extra minute of checking is far easier than trying to recover money afterward.
Frequently Asked Questions
If I can see the person's face on a video call, is it really them?
Not necessarily. The FBI says criminals use AI-generated video in real-time chats to pose as company executives, law enforcement and others, and use videos to "prove" an online contact is real. Even if you see a face, verify through a number you looked up yourself.
How should my family set up a code word?
The FBI recommends agreeing on a secret word or phrase with your family to verify identity. Pick something only your family would understand that never appears on social media, and ask anyone claiming to be a relative in trouble to say it first.
What is the penalty in Taiwan for fraud using AI deepfakes?
Under Article 339-4 of Taiwan's Criminal Code, fraud using computer synthesis or similar methods to create false images or sounds of another person carries one to seven years in prison. If NT$1 million or more is obtained, the Fraud Crime Hazard Prevention Act raises this to three to 10 years.
Who can I call if I suspect an AI scam?
In Taiwan, call the National Police Agency's 165 Anti-Fraud Hotline to check or report it. If you have already sent money or shared your login details, contact your bank's official customer service right away and file a report with the police.
Sources: FBI Internet Crime Complaint Center (IC3) – Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud (2024), MIT Media Lab – Detect Fakes research project, Criminal Investigation Bureau, National Police Agency – Press release on a deepfake romance scam ring (2026, Chinese), Laws & Regulations Database of the Republic of China (Taiwan) – Criminal Code Article 339-4 (Chinese), Laws & Regulations Database of the Republic of China (Taiwan) – Fraud Crime Hazard Prevention Act, Article 43 (amended January 21, 2026)