FBI: AI-Related Fraud Cost Americans $893 Million in 2025

The numbers finally forced the issue. For most of the Internet Crime Complaint Center's existence, AI-assisted fraud was folded into broader categories. The 2025 annual report gave it its own line: 22,364 complaints with an AI connection and more than $893 million in reported losses.

That total covers every kind of fraud with an AI link, not one scheme. Investment scams made up the largest share by far, more than $632 million, according to the FBI's 2025 IC3 report. Business email compromise involving AI added about $30 million. Distress scams, where a cloned voice imitates a loved one in trouble, accounted for just over $5 million.

Finance professor Pawan Jain, who studies fraud patterns, put it plainly: AI has made familiar scams cheaper, faster, and far harder to detect.

Where the Money Went

Investment fraud leads because AI lets scammers run thousands of tailored conversations at once. The report describes investment clubs that use AI-generated videos and voices of celebrities, executives and other trusted figures to sell fake opportunities, often with professional-looking endorsements on social media. Overall investment fraud losses topped $8 billion, which the FBI says shows many victims never realize AI was involved.

Americans over 60 reported $352 million of the AI-related total. The report does not attribute that figure to any single scheme.

The Voice-Clone Call

Distress scams are a small slice of the total, but they are the version most families picture. A scammer pulls a grandchild's voice from public social media. Voice-cloning tools can produce a convincing replica from about 15 seconds of audio. Then comes the call: a voice that sounds like the grandchild, panicked, saying he was in an accident or arrested and needs money now. Please don't tell Mom and Dad. The FBI says these scams are spreading to imitate other relatives and close friends in other kinds of emergencies.

The scam works because the urgency is designed to prevent the one thing that would stop it: taking a breath and making a second call. Studies show people often can't tell a cloned voice from a real one. Jain's advice borrows from how banks manage risk: end the call and reach the person through a number you already trust, agree on an emergency code word as a family, and require a second person to approve large transfers.

Trust Onion builds that code word into the phone. Families and teams share three codewords that rotate every 60 seconds. When a call claims to be a grandchild in trouble, the grandparent asks one question: "What are the words?" A real grandchild opens the app and reads them. A cloned voice does not come with the current words. A failed callback does not prove the original caller was legitimate, so verify through a number you already know, and use common sense if someone may be in danger.

A voice can be copied from a short clip. Three words that change every minute cannot.

A controlled voice-cloning safety demonstration

Try our tool to see just how easy voice cloning is. This controlled safety demonstration exists only to show how easy voice cloning is and to make the risk tangible. It is not a real incident or proof of anyone's identity. Your submitted sample is handled transiently to generate the fixed demonstration audio. Trust Onion does not identify you or the person from the sample. Trust Onion does not store the submitted recording, clone, or generated audio.