I spent the last few weeks pulling fraud reports, market research, and court filings for this one. I kept a running list of every stat I found, and by the end I couldn't find a single number that had gone down from the year before.

A decade ago "AI" meant robots in a movie trailer. Nobody thought about it while making breakfast. Now it unlocks your phone before your thumb even fully lands, it queues up your next video, and quietly, in the background, it's involved in decisions you never actually see happen. The price you get quoted. Whether your loan goes through. Whether a human ever opens your resume. In some cities, whether a police car ends up outside your house. None of this makes AI the villain on its own. The real question is who's got their hands on it, and how easy it is for them to misuse it.

I broke this down into three groups that can point AI at ordinary people: hackers, companies, and governments. Every figure below traces back to a police report, a university study, or a research center, and I've linked all of it at the bottom so you can check my work yourself. There's also a short list of habits at the end, things you can actually put into practice instead of just agreeing with and forgetting by tomorrow.

$893M is what Americans lost to AI-related scams in 2025, according to the FBI. 50B+ face images sit in the database of just one facial recognition company, and that number keeps climbing. $512B is roughly where analysts think the data broker industry lands by 2033, which is a lot bigger than I expected going in.

AI in the hands of hackers

Danger one

Phishing emails used to give themselves away fast. Clunky grammar, a logo that looked slightly off, a sender address that didn't quite match. Those days are mostly gone. Two tools are doing most of the damage now: deepfakes, meaning video of a real person saying or doing something that never happened, and voice cloning, which can rebuild someone's actual voice from just a handful of seconds of audio.

Case: Hong Kong, 2024

The $25 million video call

A finance employee at the engineering firm Arup joined a video call he believed included his company's CFO and several coworkers. Every person on that call was fake, generated from public footage of the real executives. He trusted it enough to wire over $25 million across fifteen separate payments. It took a week for anyone to catch it.

“Nobody on the call had been real. Not the CFO, not the colleagues. Every face he saw and every voice he heard were AI deepfakes.”

It wasn't a one-off. For the first time in the FBI's 25 years of tracking internet crime, the 2025 Internet Crime Report gave AI its own category: 22,364 complaints, about $893 million in losses. Investment scams accounted for the largest share at roughly $632 million, followed by business email compromise at about $30 million, tech support scams near $19.5 million, and romance or confidence scams close to $19 million. Investigators think the actual total is higher since most victims never realize AI was involved to begin with.

Voice cloning is the one that gets me the most, because it targets trust between family members directly. Scammers need roughly three seconds of someone's voice, lifted from a TikTok clip, a voicemail greeting, anything public, to fake a call that sounds exactly like a loved one in trouble.

AI in the hands of companies

Danger two

Most companies aren't trying to hurt anyone. They're trying to sell things efficiently, same as they always have. But AI has changed ordinary advertising into something a lot more invasive than it used to be. Recommendation engines used to be simple: show you something similar to what you just bought. Now they try to predict what you'll click, buy, or even start thinking about next, building a profile of your habits, your moods, and your weak points that never stops updating.

Behind a lot of this sits an industry most people have never heard of: data brokers. One of the biggest, Acxiom, holds records on roughly 2.5 billion people worldwide, with over 12,000 data points per person. Put together, these companies reach something like 90 percent of US Facebook users.

What gets me is how cheap this information actually is. Data on an 18-to-25-year-old has sold for as little as 36 cents. And that same data can quietly shape the price you're charged, the terms of a loan, or an insurance premium, all without you ever finding out it happened.

AI in the hands of governments

Danger three

Of the three groups here, governments have the longest reach by far. They can combine private company data with cameras, ID systems, and actual legal authority behind it. Facial recognition is the clearest example: software that scans a face and checks it against a massive photo database in seconds.

One company in particular, Clearview AI, built its entire database by scraping billions of photos off social media and news sites, no permission asked and no opt-out offered. Its collection has kept growing at a pace that's genuinely hard to picture.

Supporters argue it helps solve crimes faster, and maybe it does sometimes. But it's already hurt innocent people too. Federal testing found some facial recognition software misidentifies Black and Asian faces up to 100 times more often than white faces. At least eight Americans have been wrongfully arrested because of a bad match. Seven of them were Black.

“I knew I was innocent, so how do I beat a machine?”

The algorithm running your life

Danger four

An algorithm is just a set of rules a computer follows to make a decision. Simple enough in theory. In practice, these invisible rule sets end up shaping a surprising amount of what you see, what gets offered to you, and what you pay for things.

Social feeds rank posts to keep you scrolling, not to show you what's actually true. Pricing is already personalized in a lot of places, so two people can look at the exact same item and see two different numbers. Hiring tools have gotten it wrong too. One major tech company's AI recruiting tool taught itself to favor male applicants simply because it learned from years of mostly male resumes. Credit and insurance scoring can weigh data points you'll never see, let alone get a chance to challenge.

None of these systems announce themselves. Nobody texts you to say your price went up because of your browsing history. That's the trap. It's hard to argue with a decision you never even knew was made.

Older adults, the most targeted

Highest risk group

Every danger on this list lands hardest on older adults, and it isn't close. In 2025, Americans 60 and up reported $7.75 billion in fraud losses, a 59 percent jump in a single year. Investment schemes, fake tech support calls, and romance scams top the list, and AI voice cloning is now showing up in all three.

59% was the rise in senior fraud losses in a single year. 96% is the reported wrong-person rate when facial ID is used alone with no human double check. 3 seconds is all it takes to clone someone's voice. 4,000+ data broker companies are operating worldwide right now.

Why this danger has no safe owner

The fix seems obvious at first: keep AI out of criminals' hands, let the responsible companies and governments use it well, and call it done. But that's not the whole story. The same facial recognition software that helps find a missing kid can also misidentify an innocent adult and land them in jail. The same voice cloning tech behind a friendly customer service bot can fake a grandchild's cry for help. The same data broker pipeline that helps a bank catch fraud is what builds the cheap, detailed profile that fraudsters go and buy.

Even well-intentioned institutions make mistakes, get hacked, or let a tool creep far beyond what it was originally built for. Researchers actually have a name for that last one: scope creep. Small phrase, big problem. The point isn't that any one group here is the villain. It's that concentrated, invisible power over people's data is risky no matter who's holding it, and there's more of that power every year.

Precautions: how to actually stay safe

None of this means you need to go full paranoid. A few small habits close most of the gap.

• Set a family safe word. Use it to verify anyone calling in distress, even if the voice sounds completely real.
• Never act on urgency alone. Scammers, human or AI, always want you moving before you have time to think it through.
• Hang up and call back on a number you already had saved, never one given to you during the call itself.
• Ask a video caller to turn sideways. A lot of deepfakes still fall apart in profile view.
• Request data deletion from brokers wherever your state or country's law allows it.
• Go easy on public audio and video of yourself. It's the raw material voice cloning tools need to work.
• Actually check your privacy settings at least once. Most people never do.
• Ask if AI was involved in a loan denial, a price, or a job rejection. You have more of a right to ask than you probably think.

Sources and notes

CNN Business, Finance worker pays out 25 million dollars after video call with deepfake chief financial officer, February 2024.

Fortune, A deepfake CFO tricked the British design firm behind the Sydney Opera House, May 2024.

Cyber Helmets, $25M deepfake CFO scam on video call, April 2026.

Federal Bureau of Investigation, Internet Crime Complaint Center, 2025 Internet Crime Report, ic3.gov, April 2026.

HousingWire, FBI: seniors lost 7.75 billion dollars to cybercrime in 2025, a 59 percent jump, April 2026.

AARP, FBI report: internet crime losses hit 20.9 billion dollars, April 2026.

Grand View Research, Data Broker Market Size And Share, Industry Report, 2025.

Pew Research Center, How Americans View Data Privacy, 2023, and AI risks, opportunities, regulation, 2025.

Biometric Update, Clearview AI database growth reporting, 2023 to 2025; TIME; The Record.

American Bar Association, Police use of AI powered facial recognition technology and the risk of racial bias, 2024.

The Washington Post, Arrested by AI: police ignore standards after facial recognition matches, 2025.

Innocence Project, When artificial intelligence gets it wrong, 2024.

arXiv, Bias in text embedding models, reference to Amazon's discontinued AI hiring tool, 2024.

Scientific American, Police facial recognition technology can't tell Black people apart, 2024.

Thanks for reading, see you next time.