McAfee unveils AI Scam Detector for text, email, and video

McAfee takes on the deepfake scam era

If you have ever stared at a text from an unknown number saying "Hi, how are you?" or panicked over an email marked "termination NDA," you are not alone. AI has made digital con artists significantly smarter and terrifyingly convincing.

At CES 2025, McAfee introduced McAfee Scam Detector, an all-in-one protection layer designed to identify and intercept fraudulent messages across text messages, emails, and live video streams in real time.

Running AI locally on your PC

According to McAfee CTO Steve Grobman, modern scammers thrive in private, unmoderated channels like encrypted messaging apps, often reaching victims under the guise of urgent security notices. Another vector is video, which led McAfee to design deepfake detectors that run across any live video stream on the web.

Personalization has become the top weapon. Grobman pointed out job scams targeting people actively hunting for work on public job boards, where applicants are asked to pay out-of-pocket for "mandatory background checks." Fake missed delivery emails and AI-cloned celebrity investment pitches round out the list.

To handle the heavy scanning workload, McAfee optimized its models to run on dedicated NPUs inside modern AI PCs, falling back to CPUs and GPUs on other machines. The email protection module, previously limited to Microsoft properties in a 2024 preview, now covers Gmail and other providers. Starting this spring, the detector will roll out to existing McAfee subscribers at no extra cost.

Why it matters

Scam defense has reached a breaking point. McAfee's data shows one in three Americans fell for a text, email, or video scam over the past year, with victims spending an average of 80 hours annually just figuring out if incoming pings are legitimate. High-end deepfake hoaxes have reportedly drained victims of up to $500,000.

We think shifting these detection models to run locally on device NPUs is a clever move. It avoids the massive server bills associated with cloud inference while neatly sidestepping the privacy nightmare of routing private messages through third-party servers. For regular users, having a silent filter vetting suspicious links and synthetic video calls adds peace of mind.

However, the structural catch remains cat-and-mouse detection dynamics. Attackers will inevitably use these automated safeguards to test and refine more subtle social engineering tactics. A scanner can flag obvious red flags, but relying entirely on software to replace basic digital skepticism is still a gamble.