What are AI watermarks and content credentials?

A food package tells you what's inside and where it was made. Some photos now carry something similar. On a few sites and apps, you may see a small "cr" icon in the corner of an image. Tap it, and a panel can show which tool made the picture, when, and whether artificial intelligence (AI) had a hand in it.³⁴⁵

You'll hear two terms, and they work in different ways.

• A watermark is a hidden signal woven into the picture, the sound, or the text itself.
• A content credential is a signed record attached to the file, describing where it came from and how it changed.

One lives inside the media. The other travels with it, like the label on the package.¹²

An invisible watermark changes the content in ways people can't see or hear, but software can pick up. You'd never notice it. Google DeepMind's SynthID is one example. It adds a mark to images, audio, video, and text made with Google's own AI tools, and a matching detector can check for it later. The mark is designed to survive everyday changes, like cropping, filters, and compression. Google also says it isn't foolproof against extreme edits. And its detector looks for its own mark, so it can't speak for pictures made with other companies' tools.²

A content credential is a record of who made a file, with what tool, and what was done to it afterward. It follows a shared standard from the Coalition for Content Provenance and Authenticity (C2PA), a group of technology and media companies. A camera, an editing app, or an AI image tool can add an entry when it works on the file. Over time, the entries build into a short history.¹³

The record is digitally signed, a bit like sealing an envelope. If someone changes the picture, or edits the record after it was signed, the signature stops matching. A viewer can then show that something changed. So a credential can't be quietly rewritten without the change showing.¹

Neither tool is proof on its own. A screenshot, a re-save, or an upload to a site that strips file details can remove a content credential entirely. Watermarks travel better, but heavy editing can weaken them. And plenty of honest photos carry neither, because the camera or app that made them never added one. So a missing label tells you very little. It may only mean the record was lost along the way.¹²³

Credentials aren't only for spotting AI. Some cameras can sign a photo the moment it's taken, which can help show that a real photo is real. And a credential tells you where a file came from. It doesn't tell you whether the scene it shows is true. A signed photo of a staged scene is still staged.¹

So treat both as clues. When the icon is there, tap it. Read what the record says. When it isn't, do what you'd do anyway: look at who posted it, and whether a source you trust has the same picture. Have you spotted that small icon anywhere yet?