Google’s SynthID system was used to debunk a high-profile AI-generated hoax image, marking a rare but important victory for the system.
Earlier this week, photos surfaced online that appeared to show Kentucky Sen. Mitch McConnell in extreme pain and covered in tubes in a hospital bed. The image was widely shared on Reddit and X, but by Wednesday, respected fact-checking site Snopes had debunked the image, noting that when reviewed, the image was recorded as containing the SynthID watermark, designed by Google to identify images generated by AI.
In other words, watermarks worked as expected in the victory of anti-deepfake technology.
Senator McConnell’s health has been the subject of intense speculation since he was hospitalized following an emergency call on June 14th. Since then, he has largely disappeared from public view, prompting speculation that his health may be deteriorating. However, in this case, the evidence turned out to be completely fabricated.
Announced at Google’s 2025 I/O Developer Conference, SynthID acts as an invisible signature that is visible to the SynthID algorithm but designed to be invisible to casual observers. The signature is built into the image itself, so even if the image is screen captured on multiple platforms, like the McConnell image, the signature remains.
The main limitation of SynthID is that it can only be used if the image generation tool is an active participant in the program. Gemini models have included a watermark since the program’s inception in 2025. OpenAI joined in May 2026 as part of a broader effort to combat malicious image generation. Anthropic does not participate in the program.
Users can check whether an image contains a watermark by querying the Gemini model or by uploading the image to OpenAI’s public image verification tool.
