AI image labels and Content Credentials, explained

Why this is happening now

AI tools can now create realistic photos, videos and voices in seconds. That is fun for birthday cards and useful for designers, but it also makes it easier to fake a news photo, a celebrity endorsement or a video call. Technology companies are responding with ways to label where content came from, a field known as provenance, the same word used for the history of a painting.

Method one: Content Credentials

Content Credentials are like a signed passport attached to a photo or video. They record where the file came from, such as a particular camera or app, and what was done to it, such as cropping or an AI edit. The record is protected with a digital signature, so if someone changes it, the change can be detected.

The rules for these credentials are written by an industry group called the C2PA (the Coalition for Content Provenance and Authenticity), which includes major technology and media companies. Support is spreading:

• Google's Pixel 10 phones, launched in August 2025, add credentials to every photo taken with the built-in camera.
• In May 2026 Google said it was extending this to video recorded on Pixel 8, 9 and 10 phones.
• Some professional cameras and photo-editing software also add them.

Method two: invisible watermarks

A watermark here is a hidden pattern woven into the pixels or sound of AI-made content. You cannot see or hear it, but a checking tool can. Google's version is called SynthID, and it is added to content made by Google's AI tools.

In May 2026 Google said it was bringing SynthID checks into Google Search and Chrome, so you can select an image and ask whether it was made with AI, and that OpenAI and others were partnering to use SynthID too. When Apple released iOS 27 in September 2026, it said images made or edited with its AI tools can be identified through information stored in the image, with SynthID support due in a later update this year.

What labels cannot tell you

These tools are useful, but they have real limits, and it is worth being clear about them:

• No label does not mean real. Many cameras and apps do not add credentials, and AI tools from companies outside these schemes may add no watermark at all.
• Labels can be lost. Attached records can disappear when someone takes a screenshot or a site re-saves the file. Watermarks are designed to be harder to remove, but no method is perfect.
• A label shows origin, not truth. A genuine photo can still be shown with a false caption, out of context or from a different year.
• Rollout is uneven. At the time of writing, some announced features had no firm date for every country and platform.

There is also a risk of over-trusting the tools. If people come to believe that every fake will carry a warning, an unlabelled fake may seem more believable, not less. The healthiest attitude is to see labels as a helpful extra, like a food label that tells you some ingredients but not whether the meal was cooked safely.

How to use these tools sensibly

1. Treat a label as one clue. If a picture says "made with AI", believe it. If it has no label, keep thinking.
2. Ask the old questions. Who posted this? Where else is it reported? Does a trusted news outlet have it?
3. Use the checks where offered. If your browser or search engine lets you ask whether an image is AI-made, try it on things that matter.
4. Slow down with emotional content. Shocking, outrageous or heart-tugging pictures are the ones most worth checking before sharing.

Tip: For the bigger picture on fakes, scams and trust, read what AI means for online trust.