Mitch McConnell's hospital jeans were real. But the image of a crying Nekima Levy Armstrong, an ICE protestor held up as a symbol of "far-left agitators" by the Trump administration, was faked. 

President Trump's recent outing with Mike Tyson was legitimate. The viral photo of an adulterous Georgia city mayor was not. 

In the grand scheme of things, it has never been easy to detect AI-generated content. But it used to have its tells: a few extra fingers, or a strange texture, or an uncanny gaze. 

Now, "those tells are mostly gone, and whatever replaces them will be gone in a year too," said Marijus Briedis, chief technology officer at online security company NordVPN. "Any guide built on visual artifacts has a short shelf life."

The risk of spreading misinformation is not the only reason for questioning what you see and hear online. Digital scams, including deepfakes, led to losses of over $148 billion in 2025, an increase of more than 25 percent from the year before.

So if the visual Turing Test isn't the failsafe against fraud, scams, and manipulation we once thought it was, what do we do? We focus on the human things we can vet: Our emotions, our brains, and the digital tools available to us.

Instead of trying to spot AI mistakes, users are now encouraged by cybersecurity experts to verify sources rather than content. 

When it comes to video, don't hunt for glitches, which are being patched out of generative models fast, explains Roei Sherman, senior director of research at cloud security firm Mitiga.

Source link