STORY UPDATED: check for updates below.

Made With AI

Does a real video show inappropriate "before" behavior by a woman later photographed surrounded by Patriot Front on a Washington, DC, Metro train on July 4, 2026? No, that's not true: The video captioned "before Patriot Front" is unsourced and unverified. Google's Gemini verification tool determined the video "was edited or generated using Google AI" at the time of writing but it later changed its finding. There is a visual glitch, a missing reflection in the train window, that points to the video having been made with AI.

The 15-second video appeared in a post published by the X account @ryanasanchez on July 6, 2026 (archived here). The post was captioned:

The 'tough lady' in question:

This is a screenshot of the video:

(Image source: post by @ryanasanchez on X.)

The first eight seconds of the 15-second video show a woman kneeling and bent forward on the floor pulling down her shorts and then standing up and pulling her shorts back up. A semi-transparent black text box obscures an indecipherable social media handle (pictured above). A legible caption in white letters reads:

BEFORE PATRIOT FRONT

Featured in the opening frame, and again at nine seconds, the video shows a still image that is a real Reuters press photo taken on July 4, 2026. The caption of the photo appearing at reutersconnect.com is:

A commuter sits as members of the group Patriot Front ride the metro on the 250th anniversary of U.S. independence on the Fourth of July in Washington, D.C., U.S., July 4, 2026. REUTERS/Cheney Orr TPX IMAGES OF THE DAY

Also at nine seconds, the @ryanasanchez video shows a text box overlaid on the still image (pictured below). It reads:

AFTER PATRIOT FRONT

(Image source: post by @ryanasanchez on X.)

Lead Stories took the eight-second clip from the first portion of the video and uploaded it to Google Gemini, asking if there was a SynthID watermark. The reply (pictured below) said in part:

The video you shared was edited or generated using Google AI, as indicated by SynthID signals detected in both its visuals and audio.

Since the time of publication, Lead Stories has become aware of a glitch in Google's verification tool, discovering that it no longer produced the same positive reading for the SynthID watermark it had detected at the time of writing. Lead Stories has now published an article on our experience with this glitch and our testing to replicate it. We have shared our findings through the feedback tool in Google Gemini. It is important to understand that the absence of a SynthID watermark, a negative reading or a false reading does not mean that the video was not AI-generated.

Previously, a positive reading could be considered concrete evidence that Google had identified a SynthID watermark embedded in AI-generated content made with its own tools. Since discovering this glitch and experimenting to replicate it, the assurance of a positive reading is now questionable in certain conditions.

(Image source: gemini.google.com.)

Google's DeepMind watermark called SynthID (archived here) is detectable by their own verification tools but not visible to the human eye.

One glitch visible in the video -- typical of AI-generated flaws -- is the reflection in the Metro train window. The seats and railings are reflected, but the blue plastic bag on the seat is not reflected (pictured below).

(Image source: post by @ryanasanchez on X.)

A July 9, 2026, tip to Lead Stories pointed out an earlier copy of the video published by @symphonynocturne on Instagram on July 5, 2026 (archived here). In this copy of the video there are no captions applied in black text boxes. The social media handle that was later obscured behind the text boxes is visible -- it appears to promote an account that makes watermarks and reads "@juli watermark".