Spotting fake AI images by their flaws no longer works, experts say

Looking for flaws in a picture or video is no longer a reliable way to catch an AI fake. Security experts told Mashable that the old giveaways are mostly gone. Their advice now is to check where something came from and how it makes you feel, instead of studying the pixels.

What changed

AI content was never easy to detect, but it used to have tells, such as extra fingers, odd textures or a strange gaze. "Those tells are mostly gone, and whatever replaces them will be gone in a year too," said Marijus Briedis, chief technology officer at NordVPN. The stakes are real. Digital scams, including deepfakes, caused over $148 billion in losses in 2025, up more than 25 percent from the year before.

Real and fake images now circulate side by side. Mitch McConnell's hospital jeans were real. A crying image of protester Nekima Levy Armstrong was faked. So was a viral photo of an adulterous Georgia city mayor.

Check the source

Roei Sherman, a research director at cloud security firm Mitiga, says not to hunt for glitches, because they get patched out fast. "Every detection tip we publish eventually becomes a line in someone's prompt." Instead, look for the bigger picture. If a clip of a candidate is going around, find the full speech or interview, and look for coverage from several outlets. Short clips cut out of context are where most manipulation lives, Sherman said.

For images, a reverse image search can still help. But Sofia Rubinson of the watchdog NewsGuard warned that Google's AI-powered search often misattributes images as showing what the people spreading falsehoods claim. Briedis suggests three questions about any source:

• Has this account existed for more than a month?
• Does the same information appear elsewhere, from a familiar, credible name?
• Is this person asking me to act right now, and if so, why?

None of that requires analyzing an image, Briedis said, which is why it still works.

Ask if it makes sense

Experts also say to think about plausibility. How likely is it that this really happened? Would a public figure really say this on a public platform? Rubinson urges people to look for corroborating information and to see multiple angles of an event before believing it. Incendiary comments from politicians or controversial acts by celebrities deserve a double-check, even though many turn out to be true.

Watch your reaction

Sherman says to notice how a post makes you feel. If it enrages you or pushes you to jump in, that may be the goal. Rubinson added that one wild video showing up again and again in your feed, seemingly built to provoke an emotional response, is a reason for suspicion.

What to do

• Pause before sharing a shocking clip or acting on an urgent message.
• Find the full video or the original context.
• Look for the same story from other trusted outlets.