Watermarking

It's a marker embedded in AI-generated images, videos, and text. Rather than a visible logo, the mainstream approach embeds signals into pixel and statistical patterns that are invisible to the human eye but readable by detection software. Google's SynthID is a well-known example, and C2PA is an industry-standard format for recording a piece of content's origin and history within the file itself.

Because it's a key technology for countering deepfakes, it's closely tied to regulatory news — many countries are moving toward requiring AI-generated content to be labeled. But its limits are clear too: watermarks can sometimes be removed through cropping or recompression, and content made with open models that don't embed watermarks has none to begin with.