AI Voice and Deepfake Scams: How They Work and How to Spot Them

AI voice and deepfake scams use synthetic audio or video to impersonate someone the target trusts. That could be a relative, an executive or a government official. The scammer then presses for money, documents or login codes. The FBI says AI-generated content is now often hard to recognize. Both it and the Federal Trade Commission (FTC) advise verifying an unexpected request independently before acting.

How do AI voice scams work?

Cloning a voice takes little more than voice-cloning software and a brief recording of the person speaking, according to the FTC. The agency says the recording can be lifted from something they posted online. The clone is then used in what the agency calls a family emergency scam. That is a call that sounds like a relative who has crashed a car or been arrested and needs money. The caller says the matter is urgent and may ask the target to keep it secret.

Officials and their contacts are targeted too. The FBI warned in May 2025, and again in December 2025, about people posing as senior US officials. It said they were sending text messages and AI-generated voice messages to build rapport with targets. The December alert said the activity dates back to 2023. According to that alert, the impersonators asked to move the conversation to apps such as Signal, Telegram or WhatsApp. It said they also requested authentication codes, personal documents or wire transfers.

How do deepfake video scams work?

Deepfakes are AI-generated videos that imitate real people. In one widely reported case, the British design firm Arup confirmed in 2024 that fake voices and images had been used against it. An employee in its Hong Kong office sent $25.6 million (200 million Hong Kong dollars) to fraudsters across 15 transactions, Fortune reported. Fortune said the payments followed a video call featuring deepfakes of the chief financial officer and other staff.

The largest AI-linked losses reported to the FBI involve investment pitches. The bureau's Internet Crime Complaint Center (IC3) says fraudulent investment clubs use AI-generated videos and voices of celebrities and chief executives as fake endorsements. Its report for 2025 counted 22,364 complaints that referred to AI, with losses above $893 million. More than $632 million of that came from investment scams. The report separately put losses to "distress" scams at more than $5 million. In those scams, a cloned voice mimics a loved one. The report also said many victims are unaware of how far AI may be involved.

Scammers also target people who have already been defrauded. The FBI warned in July 2026 about criminals posing as IC3. It said they were using deepfake videos of FBI personnel to steer victims to imitation IC3 websites. The bureau said IC3 never contacts individuals directly by phone, email or social media. It also said IC3 never asks for payment to recover lost funds.

How to spot AI voice scams

The request is a better guide than the sound of the voice. Warning signs listed by the FTC and FBI include:

• Pressure to act immediately and to keep the request secret.
• Payment by wire transfer, cryptocurrency or gift cards.
• A request for a verification code, or to switch to a different messaging app.
• Tone or word choice that does not fit the person.

Hearing alone is not dependable. In a 2023 study published in PLOS ONE, University College London researchers found that 529 participants correctly identified deepfake speech only 73% of the time. They also found that playing them examples beforehand improved accuracy only slightly.

Both agencies recommend ending the call and contacting the person or organization directly. They say to use a number you already know or have looked up yourself. A December 2024 FBI alert also suggests agreeing on a secret word or phrase with family members to confirm identity.

How to tell if a video is fake

The same alert lists possible giveaways in AI-generated images and video:

• warped hands or feet
• teeth or eyes that look wrong
• blurred or uneven faces
• odd-looking glasses or jewelry
• shadows that fall the wrong way
• lag and unnatural movement

In job interviews that use voice deepfakes, IC3 says, lip movements may not match the audio. It says coughs or sneezes may not match the picture.

The FBI's July 2026 alert said AI-generated material is now often difficult to recognize. It said a cloned voice can sound almost the same as the real one. A clean-looking video is therefore not proof of authenticity. For any payment or data request, the safer test is to end the call and confirm through a contact route set up beforehand.

The bottom line

Imposter scams were the top scam category reported to the FTC in 2025, the agency said in May 2026. That made it the ninth year in a row, the FTC said. It counted more than 1 million such reports. Reported losses rose by nearly 20% to $3.5 billion, it said. That figure covers all imposter scams, not only those that use AI.

The practical rule is to verify before sending money, codes or documents, however familiar the voice or face. The FBI also advises limiting public recordings of one's voice and image. Suspected scams can be reported to the FTC at ReportFraud.ftc.gov and to the FBI at ic3.gov.