How to Protect Yourself Against AI Scams

In 2024, a series of ads on Facebook and other social media sites appeared to show Taylor Swift hawking free Le Creuset cookware. Except, as the New York Times reported, it wasn’t really Swift but a deepfake that looked and sounded like her. Scammers had used artificial intelligence to synthesize the singer’s voice and combined it with video footage of her to create a credible doppelganger. 

Fans who responded to the ad were sent to websites that looked legitimate. But when they placed orders there, they were hit with fraudulent shipping fees and hidden monthly service charges, and their personal data was compromised. And the cookware never arrived.

As AI becomes more powerful and more accessible, cybercriminals are using it to create scams that are more personalized and harder to detect—all at a scale never possible before. “AI is increasing the velocity and believability of scams,” says Abhishek Karnik, director of threat research and response at digital security firm McAfee.

AI scams spread.

It’s not just consumers who are getting ripped off. That same year, an employee working in Hong Kong for the U.K. global consultancy firm Arup was duped into transferring more than $25 million to fraudsters who used deepfake technology to impersonate the company’s chief financial officer and other employees in a video conference call. 

Large companies that are integrating AI into their operations are also falling prey to scammers. Almost every Fortune 500 firm has since 2024 unwittingly hired remote North Korean tech workers, according to Axios. Many of these candidates used AI to disguise their true location and identities and to pass interviews and background checks. 

According to the FBI, the North Korean government allegedly deployed these workers to surreptitiously get businesses around the world to hire them, as a way of generating revenue. Some of those workers also “introduced malware into company networks to exfiltrate proprietary and sensitive data,” according to the Treasury Department. 

Evolution of AI

To understand how AI is empowering bad actors, it helps first to understand how AI has been evolving.

Early AI models analyzed existing text or data to make decisions, predictions, or recommendations. That in turn led to big improvements in things like online search. “We could ask questions and get better, faster answers,” says Gil Baram, a research scholar with the Center for Long-Term Cybersecurity at University of California Berkeley. 

Generative AI

Then came generative AI, which could create new content. It’s the brains behind such popular programs as OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini. “Gen AI has allowed us to be more creative in creating text, audio, and video,” Karnik said. At the same time, though, “the scammers are also getting more creative.”

For example, McAfee first started seeing toll scams in 2024: People would get a text from a legitimate-looking toll authority (such as FasTrak in California) claiming they had unpaid tolls and would face fines or account suspension if they didn’t make an immediate payment. The scammers collected any payments made to the fraudulent websites. Such bogus texts were nothing new. But generative AI enabled scammers to personalize and produce them at huge scale.

Criminals often gain the trust of victims through social engineering. They reach out via phone calls, texts, or emails that seem to be from family, friends, co-workers, legitimate companies, or government agencies. They want you to click on a link, download an app, or open an attachment that leads to a fraudulent website or infects your computer or device with malicious software.

Historically, such scams could often be spotted by bad spelling, awkward grammar, or foreign accents. With generative AI, scammers can write text like an English teacher and sound like a native-born American. “We used to say, if audio sounds like a robot, it’s fake,” says Karnik. “Now, it stutters, it takes a moment to breathe—it sounds very real.”

Agentic AI

The next frontier, called agentic AI, creates more potential for mayhem if it’s in the wrong hands. It has agency, in the sense that it can think on its own and perform tasks, says Michael Bruemmer, Experian’s head of global data breach resolution.

Baram explains the evolution like this: While generative AI can help you plan your next vacation—you can enter your preferred dates and destination into ChatGPT, and it will recommend an itinerary, lodging, and transportation options—agentic AI can also book your flight and hotel. 

Because agentic AI systems can think, plan, and act independently, they can be “the perfect digital accomplices for sophisticated scammers,” McAfee warns.