Was That AI or Just a Bad Video Edit?

Trump’s hand broke the internet — here’s how to tell what’s real when everything looks fake.

Author

Shira Lazar
September 14, 2025

If the past week felt unreal, you’re not alone. In the aftermath of Charlie Kirk’s killing in Utah, rumors and synthetic spin outpaced facts at every turn. Then came President Trump’s response video, where, for a split second, his left pinky seems to “fuse” with the rest of his hand. That micro-glitch was enough to launch a thousand threads insisting the White House deepfaked the president. Except…video and forensics experts say there’s no evidence it was AI at all, just heavy color grading, compression, and social-platform re-encoding, turning a small video glitch into a conspiracy.

Meanwhile, real misinformation did flood the zone. AI chatbots and viral posts misidentified suspects and mangled basic facts as the investigation unfolded.

So how do we check reality now—for real?

@shiralazar

🚨 AI-generated fake images are taking over—but do you know how to spot them? 🤔 Here’s how to outsmart the fake: 👋 Look for weird hands. 🎨 ... See more

The tools that actually help today

Here are three categories you’ll see in the wild—and how to use them without getting burned.

1) Content Credentials (C2PA) — provenance, not detection

Think of this as a tamper-evident “nutrition label” attached by cameras and editing software. When present, you can verify the capture device, edits, and whether AI-assisted. Major news orgs and platforms are piloting it; some cameras (Leica, Nikon, Sony) and YouTube workflows already support it.

2) Google DeepMind’s SynthID — watermark detection for Google-made media

If an image or (increasingly) audio/video was generated by Google’s models (Gemini/Imagen/Veo), SynthID can embed and detect an invisible watermark. Google’s public SynthID Detector lets you check files for those marks; helpful in platform workflows, limited for non-Google models. (Google DeepMind)

3) Enterprise deepfake detectors (e.g., Reality Defender) — “best effort” classifiers

These services score likelihood that media is synthetic and are used in risk-sensitive environments like banks and call centers. They’re valuable signals, especially for audio fraud—but still fall to the generalization gap that researchers warn about. Treat results as one piece of a multi-method investigation, not a final ruling. 

The Trump clip wasn’t AI. But the fact that so many of us instantly thought it was, and that half the “AI detectors” out there said the same, should stop us in our tracks. It shows how fragile trust has become. 

While there are some apps and software out there claiming to be AI detectors and verifiers, they are still new and not perfect. A lot of them are actually using AI for the detection, so we really can’t trust them yet. So much of this leaves us with more questions than answers. 

Something to think about through all of this is also the fact of what happens when people are using AI to generate or edit their videos. 

 As politicians begin experimenting with “AI digital clones” for public announcements, how will we know what is real? And if it is done with consent, does the distinction even matter?

My takeaway:The bigger issue isn’t only about detection. It’s about maintaining a standard of presence and accountability when the stakes are highest. Just as you wouldn’t deliver devastating personal news over text, we should expect our leaders — in politics, business, and culture — to show up as people when addressing matters that shape our safety and future. Anything less risks weakening the trust we all depend on.

Shira Lazar

Other headlines to check out:

AI

Creator Economy

Web3 

🎧 New Episode of The AI Download: AI, Sex & the “Technosexual” Era: Kaamna Bhojwani on Love, Shame, and Robot Romance

This week on The AI Download, I had one of the most fascinating (and honestly fun) conversations yet. With sexologist and Psychology Today columnist Kaamna Bhojwani. We talked about AI boyfriends, VR bedrooms, and why so many of us still carry shame when it comes to intimacy.

Kaamna broke down what it really means to be “technosexual,” why some people say an AI companion is better than nothing, and how tech can actually help us practice the stuff that makes relationships last—communication, creativity, and yes…cunnilingus.

🎙️ Listen now to get caught up on the AI headlines you need to know plus insights on future of life, creativity and culture. 

Gentle Reminder 🙏

"The way to get started is to quit talking and begin doing" - Walt Disney

Advertise with Us

Remember, I'm Bullish on you!

With gratitude,