AI watermark
Also called SynthID, watermarking, invisible watermark
A hidden signal built into AI-generated images, video, audio or text so software can later tell it was made by AI.
A visible watermark is a logo in the corner, easy to crop. An AI watermark is woven into the content itself: tiny patterns in an image's pixels, an audio track or a video that people can't notice but software can detect. For text, Google's method nudges which words the model picks, in a pattern a detector can recognise.
Watermarks are one of the main answers AI companies have offered to deepfakes. Google's SynthID is the most widespread; ChatGPT images carry it too, alongside Content Credentials, which are signed labels stored in the file's data.
They help, within limits. A watermark proves where something came from only if the maker added one and nobody stripped it out, and open models anyone can run don't have to add anything. Treat a found watermark as useful evidence and a missing one as no evidence at all.
For example
You upload a suspicious photo to a checker, which finds Google's SynthID watermark and says it was made with Google's AI.
Put it to use
Where you'll meet AI watermark in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.