of Google SynthID The system has been used to debunk a high-profile hoax image created by AI, in a rare but significant win for the system.
Earlier this week, a photo surfaced online that appeared to show Kentucky Senator Mitch McConnell covered in tubes in a hospital bed in extreme distress. The image was shared widely Reddit and Xbut by Wednesday, the respected fact-checking website Snopes had debunked the image, noting that when selected, the image is listed as containing the SynthID watermark designed by Google to identify AI-generated images.
In short, the watermark worked exactly as it should in a win for anti-deepfake technology.
Senator McConnell’s health has been the subject of intense speculation since he entered hospital after an emergency call on June 14. Since then, he has been largely absent from the public eye, fueling speculation that his health may be deteriorating. In this case, however, the evidence turned out to be completely bogus.
Launched at Google’s 2025 I/O developer conference, SynthID works as an invisible signature, visible to SynthID algorithms but designed to be invisible to the casual observer. Because the signature is embedded in the image itself, it survives even when an image is recorded on multiple platforms, as the McConnell image was.
The main limitation of SynthID is that it can only be used when an imaging tool is actively involved in the program. Gemini models have included the watermark since the program began in 2025. OpenAI joined in May 2026, as part of a broader effort to combat the creation of malicious images. Anthropic does not participate in the program.
Users can check if images contain the watermark by asking a Gemini model or uploading it to OpenAI’s public image verification tool.
