Google’s SynthID watermark flags AI-generated image of Senator Mitch McConnell

A manipulated image purporting to show Senator Mitch McConnell in a hospital bed with medical tubes was widely shared on Reddit and X earlier this week. The image was later debunked as AI-generated after fact-checkers at Snopes analyzed it and found it contained a SynthID watermark, a signature embedded by Google’s tool designed to identify AI-generated images.

Google’s SynthID system, launched at the company’s I/O developer conference in 2025, embeds an invisible watermark directly into AI-generated images. This watermark is detectable by SynthID algorithms but remains unnoticeable to casual observers. Because the watermark is part of the image file itself, it persists even when the image is screencaptured and redistributed across platforms, as occurred with the McConnell image.

The SynthID watermark system is currently operational only when the image-generation tool participates in the program. Google’s Gemini models have included the watermark since the program’s launch in 2025. OpenAI joined the initiative in May 2026 as part of a broader effort to combat malicious image generation. Anthropic does not participate in the program.

Users can verify whether an image contains a SynthID watermark by asking a Gemini model or uploading the image to OpenAI’s public image verification tool. The McConnell image debunking demonstrates a rare but significant real-world validation of the system’s effectiveness in flagging AI-generated content.