Top Deepfakes of the Week: The AI Fakes That Fooled the Internet (July 20–26, 2026)

Deepfakes This Week: The Full Rundown

Every week, something AI-generated goes viral before it gets caught. We track it so you don't have to find out the hard way — usually after you've already shared it. Here's the countdown, worst first.

#3 — Scammers Deepfaked the FBI to Rob People Who'd Already Been Robbed

This one is genuinely unsettling. On July 20, the FBI's Internet Crime Complaint Center (IC3) put out a warning: scammers are now using AI-generated video of fake IC3 officials, paired with a cloned version of the real IC3 website, to contact people who have already lost money to fraud once.

Think about the target here. Not a random stranger — someone who just got scammed, is scared, and is actively looking for help. That's exactly who clicks "official" looking video from someone claiming to be the agency that's supposed to protect them.

It's the same trick as classic recovery scams, except now the "official" on the call looks and sounds like a real federal agent. If you've ever wondered why deepfake detection for scam and fraud content has become its own category of tool, this is exactly why.

#2 — A Prime Minister Got Deepfaked Threatening His Own Citizens

In Delhi, as protests over a paper-leak scandal escalated, two separate AI-doctored videos started circulating within a day of each other. One showed Prime Minister Narendra Modi appearing to threaten protesting students. The other showed a senior cabinet minister, Piyush Goyal, doing the same thing outside Parliament.

Both were built the same way: real footage, swapped audio. The visuals are untouched, so there's nothing "off" to spot by eye — the mouth movement lines up perfectly, because it's real. Only the words are fake. India's PIB Fact Check unit called both out publicly the same day they surfaced, as News24 reported.

Goyal didn't just deny it. He filed a police complaint, and an FIR was registered within hours, according to Organiser's report. That's a strikingly fast institutional response compared to most countries — most fakes are still circulating long after the person they target even finds out.

#1 — The Trump "Collapse" Video That Fooled 4 Million People (and Had a Hidden Disclaimer)

This is the one everyone was talking about. On July 24, a video appeared to show President Trump, 80, collapsing while getting into an SUV. It spread across X and TikTok, with one version alone racking up more than 4 million views. People immediately treated it as evidence of a health crisis.

Here's the twist that makes this one worth remembering. The video wasn't secretly fake — the creator had labeled it as AI-generated in the caption, as Snopes documented. The catch is that the label was buried underneath a long, taunting message about the president's mortality, visible only if a viewer clicked "more." Almost nobody did. They just watched, believed, and shared.

The creator later admitted, in comments to Lead Stories, that he wanted to test how many people would "just jump" at it. A lot of people jumped. Multiple independent tools, including ZeroGPT, flagged the clip as AI-generated with high confidence once someone actually looked. Before that, it just looked like news.

This is the story that should worry you most, and not because it was political. It's because "the label was technically there" turned out to mean almost nothing once it got tucked behind a click. That gap — between disclosed and actually seen — is where most viral deepfakes live now, whether it's a manipulated video of a public figure or a fabricated voice clip of someone you know.

Why This Keeps Working

How to Actually Catch One of These Yourself

None of this week's fakes were hard to catch with the right tools — they were only hard to catch by eye, in the ten seconds most people spend scrolling past a video before sharing it. That's the entire problem in one sentence.

If a video, image, or voice clip made you feel something strong and fast — outrage, fear, urgency — that's usually the moment to slow down and check it, not share it. Run it through a proper AI image detector or video scanner before you decide it's real. It takes less time than writing the caption you were about to add.