What happened
An AI-generated image showing Senator Mitch McConnell in a hospital bed was debunked this week after circulating widely on social media. According to TechCrunch, the fact-checking site Snopes confirmed the image was a fake by detecting Google's SynthID watermark.
The invisible signature, designed to identify AI-generated content, worked as intended. It persisted even after the image was screencaptured and re-shared across platforms like Reddit and X.
How the room's reading it
This is being framed as a rare, clear-cut win for deepfake detection. Trust and safety specialists see it as proof that technical watermarking can work in the wild, surviving the compression and re-sharing that typically degrades such signals. The conversation on X, however, is more sceptical. Developers immediately highlighted the system's main limitation — it's an opt-in programme.
While Google and OpenAI participate, many popular open-source models and other major labs do not. The consensus among practitioners is that without universal adoption, SynthID remains a partial solution for a problem that requires a comprehensive defence.
Sailfish's take
We see this less as a victory and more as a reminder of the system's fragility. The fact that this debunking is newsworthy shows how rarely these systems actually land a public win. For builders, the lesson is clear — you can't outsource trust. We've shipped enough products with user-generated content to know that relying on upstream watermarking is a losing strategy.
The responsibility for identifying synthetic media falls on the application, not the model provider. If you're allowing image uploads, this is the week to start building your own detection and moderation queues. Don't wait for the labs to solve this for you.