The fake photo carried the clue that exposed it.
An AI-generated image falsely depicting Sen. Mitch McConnell in a hospital bed was widely shared on social media before Snopes fact-checkers identified a Google SynthID watermark embedded in it. The case offers an unusual real-world demonstration of how invisible AI provenance signals could help verify suspicious images after they go viral.
It also reveals the technology’s central weakness: Watermarks can only identify content when the image generator embeds one… and the detection tool knows how to find it.
What caused the viral hoax?
According to TechCrunch, the hoax began with an AI-generated image falsely showing Senator Mitch McConnell in a hospital bed connected to medical tubes. The image spread rapidly across Reddit and X, where it gained traction and speculation about the senator’s health following his June 14 hospitalization and his limited public appearances.
McConnell’s recent medical visit gave the fabricated image a credibility many viral hoaxes lack. Rather than emerging in isolation, the fake photo capitalized on an existing news event, giving the misleading information scale.
However, the same image that convinced many users also contained the invisible evidence that ultimately exposed it.
How a SynthID technology exposed the fake image
Fact-checkers at Snopes on Wednesday, July 8, determined that AI had generated the hospital image after detecting Google’s SynthID watermark embedded within it.
Unlike traditional metadata or visible watermarks, SynthID inserts an imperceptible digital signature directly into supported AI-generated media, allowing compatible detection tools to identify their origin even after common modifications such as lossy compression, resizing, or screenshots.
The image watermarking technology was unveiled at the 2025 Google I/O developer conference and, in this case, performed exactly as Google said it would.
However, SynthID has clear limits. The technology can only identify images produced by AI systems participating in the SynthID program, so content generated by unsupported models will not be detected. TechCrunch notes that OpenAI joined the initiative in May 2026.
I also conducted a small test of both companies’ verification tools. Gemini successfully identified a cropped image generated by Google’s own AI model. However, the Google chatbot did not detect the image generated by ChatGPT.
OpenAI’s Verify tool detected an original ChatGPT-generated image but failed to recognize a screenshot of that same image or images generated by Gemini. Although the sample size was small, the results suggest that users may see different outcomes depending on which verification tool they use and how the image is modified.
What does this mean for internet users?
Sen. McConnell’s case is just one of many episodes of AI-generated media used to push false narratives online. As AI-generated images become more realistic and easily accessible, thanks to Google and other image-generation models, combating their use for malicious purposes becomes crucial.
For users, a technology like this is a win against false narratives and impersonation. To verify whether media is AI-generated, you can ask a Gemini model or upload it to OpenAI’s verification platform. Detection should be combined with source checks, reverse-image searches, and reporting from trusted outlets.
In Other News: Google Photos can now turn everyday videos into AI-generated cinematic clips. Learn how the new Video Remix feature works.