FBI: AI Voice Scams Cost Americans $893 Million in 2025

A New Category for a New Kind of Crime

The FBI doesn't add categories lightly. The Internet Crime Complaint Center has tracked online fraud since 2000, and for most of that time, AI-powered scams were folded into broader fraud buckets. Not anymore.

The 2025 annual report gave AI fraud its own line: 22,364 complaints and $893 million in losses. Those numbers reflect only what was reported. The real figure is almost certainly higher.

Finance professor Pawan Jain, who studies fraud patterns, put it plainly: AI has made familiar scams cheaper, faster, and far harder to detect. The technology didn't invent the grandson scam. It just made the scam easy to run at scale.

How the Grandson Scam Works

The setup is simple. A scammer finds a few seconds of audio from a young person's social media account, a TikTok, a YouTube video, a public Instagram reel. That's enough. Voice cloning tools can build a convincing replica from as little as three seconds of audio.

Then comes the call. An elderly grandparent picks up the phone and hears a voice that sounds exactly like their grandson. The voice is panicked. He was in a car accident. He's been arrested. He needs bail money right now. Please don't tell Mom and Dad.

A second voice gets on the line, someone posing as a lawyer or a police officer, to add authority and urgency. The grandparent is told to wire money, buy gift cards, or hand cash to a courier who will arrive at their door. The whole call can be over in ten minutes, and the money is gone before anyone realizes what happened.

Why It Works So Well on This Generation

People over 60 lost $352 million to AI-assisted fraud in 2025 alone. That's not because older adults are less intelligent. This scam is engineered to bypass logic.

The voice triggers an immediate emotional response because it sounds like someone they love. The scenario activates every protective instinct a grandparent has, and the urgency is designed to prevent the one thing that would stop the scam: taking a breath and making a second call.

Scammers explicitly tell victims not to call other family members, framing it as protecting the grandchild from embarrassment. They create a closed loop where the only information the victim receives comes from the scammer. Human detection of cloned voices isn't reliable either. Research suggests people correctly identify a fake voice only about 50 to 60 percent of the time, and that accuracy drops when the listener is emotionally activated.

The Economics Changed Everything

Voice cloning used to require expensive software and technical expertise. That barrier is gone. Today, free and low-cost tools can produce convincing audio clones in minutes, so a scammer no longer needs technical sophistication. They need a phone number and a few seconds of public audio.

This is why the FBI's numbers jumped enough to warrant a new category. The scam didn't change. The cost to run it collapsed.

Professor Jain and other researchers note that AI has also sped up the targeting process. Automated tools can scan public social media, identify likely targets (grandparents with grandchildren who post publicly), match voices to family relationships, and generate calls at a volume no human operation could match.

What Families Are Trying

Some families have started warning elderly relatives in advance: "If someone calls saying it's me and there's an emergency, call Mom first." That's good instinct, but it relies on the grandparent remembering the instruction in a moment of stress, which is exactly when memory and judgment are hardest to access.

Others have set up code words, a single agreed-upon phrase that a real family member would know. That works until someone forgets the word, or too much time passes and no one remembers if it was "pineapple" or "sunflower."

The problem with most manual solutions is that they depend on the person being scammed to execute a plan calmly in the middle of a crisis.

Three Words the Scammer Can't Know

Trust Onion was built for exactly this problem. It gives families a set of three rotating codewords that change every 60 seconds, calculated locally on each family member's phone without touching any server. When a call comes in claiming to be a grandchild in trouble, the grandparent asks one question: "What are the words?"

A real grandchild opens the app and reads them. The scammer, no matter how good the voice clone, cannot answer. They don't have the words, and nobody outside the family does. The words rotate on a schedule, so even if a word were somehow overheard, it would be useless within a minute. There's nothing to steal, nothing to intercept, nothing to fake.

AI can clone a voice in seconds. It cannot guess a codeword it was never given access to. Trust Onion is free, works offline, and your grandparents don't need to understand how it works. They just need to know one question to ask.

What to Do Right Now

Before you close this tab, think about the people in your family most likely to get this call. Statistically, someone over 60, someone who loves their grandchildren and would do anything for them in a crisis.

The FBI's new category is a signal. This is a documented, organized form of fraud that specifically targets the protective instincts of older adults. Three rotating codewords change the calculation entirely. The scammer needs a convincing voice. Your family only needs to ask one question they cannot answer.