In another part of the world, a widow thought she had found love again—online. A woman had sent images and videos of herself, with talk of spending the rest of their lives together. First, though, she had to finish her job on an ocean-based oil rig and couldn’t afford to fly out to visit. The widow kept sending money until her bank told her it was a scam.

Neither of these scams are new. What’s new is the level of sophistication—thanks to artificial intelligence—that criminals use to create seemingly real voices or images. And the problem is about to get much worse.

AI used for fraudulent reasons has become so frequent that the FBI issued a warning in December cautioning the public to not automatically trust something that might seem real, even to our own eyes. The AI enhancements to these scams are hard to spot, even by experts.

“Before [2024], we hadn’t seen one decent AI scam,” says David McClellan, founder and CEO of Social Catfish. “There had been some tools you could download and have AI deepfake-ish type videos generated, but they were very fake and not believable. From March [2024], we’ve seen a huge surge in deepfakes.”

When it comes to the AI tools used to get that done, criminals tend to follow one of three paths:

1. VOICE DUPLICATION. Using a recording of someone’s voice that’s just a few seconds long, a crook can use AI to replicate that voice almost perfectly—often well enough to deceive a loved one. That makes this tool especially popular for those running a grandparent scam. “The way to prevent that is for every family to have a code word,” says Haywood Talcove, CEO of government business at LexisNexis Risk Solutions. Unless the code word is used, “hang up and then call your relative. I guarantee they’re not going to know what you’re talking about.”

2. PHONY PHOTOS. Not long ago, scammers would scour the internet to find random photos they could use as part of their scam. Now they rely on AI to generate hundreds of images to seemingly corroborate whatever tale they might use to get their hooks in victims. A recent survey found that 72 percent of Americans couldn’t tell the difference between an AI image and a real one—and the number of wrong guesses increased among older people.

3. VIDEO DEEPFAKES. As AI gets more advanced, it’s capable of creating not just more realistic still images, but also videos. These deepfake videos can be so convincing that some people might think they’re having a video call with a real person.

Keeping pace with the speed of this technology is challenging, even for experts.

“Realistically, even as soon as midyear, we may not be able to detect this stuff, because it’s getting that good,” McClellan says.