Fact-checkers at Snopes recently debunked a fake image showing Kentucky Senator Mitch McConnell in a hospital bed by finding a hidden watermark embedded in the photo. The image had been shared widely on Reddit and X before Snopes identified it as artificially created using a tool called SynthID, made by Google TechCrunch.

McConnell had checked into a hospital in mid-June and had not been seen publicly for weeks. That absence created an opening for the fake image: a senator already known to have health concerns, no recent appearances to contradict a dramatic photo, and a hospital-bed picture that was easy to believe at first glance.

What Is SynthID?

SynthID is Google's watermarking system for AI-generated images. Think of it like an invisible stamp that marks a photo created by artificial intelligence. The watermark is embedded at the pixel level—deep within the image itself—so that it remains hidden to the human eye but survives repeated copying, compression, and sharing across social media.

The key advantage here is durability. When the McConnell image spread across Reddit and X, it was reposted, screenshot, and compressed many times. Most invisible marks would degrade and disappear. SynthID's watermark held up through all of that, which is what allowed Snopes to identify it as artificially made.

Google launched SynthID in 2025 and built it into Gemini, the company's AI image generator. Since then, OpenAI (maker of ChatGPT) joined the program in May 2026 and now also marks its AI images with the same watermark. OpenAI has even created a public tool that anyone can use to check whether an image carries the watermark TechCrunch. This means regular people, not just fact-checkers, can upload a suspicious photo and find out if it came from these companies' AI systems.

The Blind Spot

There is a problem worth understanding. The watermark only works if the AI system that created the image actually added it. Other companies—like Anthropic, which makes Claude—have not joined the watermarking program yet. That leaves room for someone to generate a fake image using a non-participating AI system or an older model without the watermark, and the detection tool would not catch it.

This matters because the McConnell image happened to carry a traceable signature. But if someone created a fake photo using a different AI system or stripped the watermark away intentionally, the verification tool would find nothing. Finding no watermark does not automatically mean an image is real; it could simply mean it came from a system that does not use watermarks.

What Changes Now

Fact-checkers like Snopes have traditionally caught fake images by examining them closely for visual clues—blurry fingers, weird backgrounds, inconsistent lighting—or by tracing where they came from. The McConnell case showed that they now have a new method: checking for the SynthID watermark. That is a meaningful shift in how misinformation gets caught and debunked.

The real test ahead is what happens when someone deliberately tries to make a fake image that defeats the watermark. That could mean using a non-participating AI system, running an image through a process designed to scrub the watermark, or using an older open-source model that never added one in the first place. So far, at least in cases that have become public, that has not occurred in a way that made major headlines.

Speed Still Matters

SynthID does not stop someone from creating a fake photo. It gives fact-checkers—and now everyday people with the OpenAI tool—a way to verify whether an image is artificially made before they share it or believe it. But that verification only works if someone actually checks the image early. If a fake photo spreads for days before fact-checkers see it, the damage is already done. In this case, Snopes caught the McConnell image relatively quickly. That will not always be the case.