A sitting president just posted an AI-edited image as proof of real-world work. That should worry anyone who relies on official photos to tell them what actually happened.
According to The Verge AI, President Donald Trump shared a set of photos on Truth Social to show off renovations at the White House. One of them, a before-and-after comparison of a White House column, appears to have been modified with AI. When The Verge uploaded it to Google’s SynthID Detector, the tool flagged it as “made or edited with OpenAI.”
🔍 What gave it away
Reddit users spotted the problems first. The Verge AI lists several discrepancies in the “restored” image:
- Steps that don’t add up. The new column has a different number of steps on each side, and neither count matches the “before” photo.
- A suspiciously identical setup. Both shots share nearly the same angle and detail. The reflections in the windows barely change.
- Cardboard that didn’t move. The protective cardboard at the base of the column seems to sit in exactly the same spot in both photos.
- Too-perfect paint. The “newly painted” column looks unnaturally smooth.
In fairness, the column may really have been painted. Real photos show workers painting the front of the White House in June. The White House didn’t immediately respond to The Verge’s request for comment. SynthID also didn’t detect a watermark on Trump’s other photos, which show the facade and the heliport.
⚠️ Why this matters
What stands out here is the kind of image. This isn’t an obvious meme or a cartoon. It’s a documentary-style photo presented as evidence that work was done. That’s the category where AI editing does the most damage, because viewers have no reason to question it.
It also fits a pattern. As The Verge AI notes, Trump often posts AI-generated caricatures of himself. The White House has posted racist AI memes and at least one photo deceptively edited to make a protester look distraught. Each case makes the next one feel more normal. Once official channels routinely blend real and synthetic images, the public loses a shared baseline for what counts as a real photo.
🛡️ The detection angle
This story is also a real-world test of watermark detection, and the result is mixed but encouraging.
SynthID is Google’s invisible watermarking system. It embeds a signal into images, video, and audio that people can’t see but software can read later. Google’s detector scans uploads for that signal and reports whether content was “created or edited” with AI.
Here’s what this case shows:
- It worked. The watermark survived being posted to a social platform and still triggered a match.
- It’s selective. It flagged one image and cleared the others, which is what you’d want from a tool like this.
- It has limits. Watermarks only catch content from tools that embed them. An image edited with a tool that doesn’t use SynthID, or one that’s been heavily altered, can slip through without a trace.
So a clean scan doesn’t prove a photo is real. A positive scan is strong evidence it isn’t.
🧭 What you should do
If you work in media, comms, or anything that depends on visual evidence, treat this as a prompt to tighten your process:
- Run high-stakes images through detection tools like SynthID Detector before you publish or share them.
- Check the physical details. Step counts, reflections, and objects that don’t move between “before” and “after” shots are classic giveaways.
- Look for corroboration. Independent photos, like the June painting shots, can confirm or undercut a claim.
- Don’t treat a clean scan as a clean bill of health. Absence of a watermark isn’t proof of authenticity.
What comes next
Expect more of this, not less. Image editing tools keep getting easier to use, and political accounts have every incentive to polish their wins. The open question is whether detection tools like SynthID become standard practice for newsrooms and platforms, or stay something only diligent reporters bother to use. Full details and the image comparisons are available in The Verge’s original report.