What this is about

The EU AI Act requires that an image generated by AI is recognisable as such. The rule applies from 2 August 2026.

What that means for you in a specific case depends on who you are and what you publish - that is a question for your legal advisers, not for a tool. What we can tell you is what kontiqo does.

Two kinds of labelling

The visible one

A note in the image or on the post that a human reads. kontiqo asks you once, on your first post with an AI image, whether you want it.

Nothing pre-ticked. The box is unchecked when you see it. You decide, and you can change that decision in settings at any time.

For screen readers too. If an image carries the label, the note also goes into the image description. Anyone using a screen reader gets it - otherwise the labelling would be invisible to exactly the people who need it most.

The machine-readable one

Every image generated in kontiqo carries the marker "trainedAlgorithmicMedia" in its metadata. That is a standardised field saying: this image came from a trained algorithm.

We have been setting this marker since June 2026 on every image, regardless of how you decided on the visible label. It does not sit in the picture but in the file, so it does not disturb the image.

What happens to the marker on upload - measured, not assumed

Here comes the uncomfortable part, and we would rather say it ourselves.

The platforms re-encode uploaded images, and metadata can be lost in the process. We checked real posts to see what arrives:

  • Facebook and Instagram: the marker survives.
  • LinkedIn: it does not.

That is not a fault of ours and not one we can fix - it happens at the receiving end. But it means: anyone downloading an image you published on LinkedIn and inspecting its metadata will no longer find the marker there.

The file in your kontiqo library still carries it, and that is the file you pass on, archive or produce on request.

This is why the visible label is not decoration: it is the labelling that survives the journey through the platform.

What we do not claim

kontiqo does not make you automatically compliant. A tool can set a marker; whether your publication as a whole meets the requirements depends on your situation.

And a second point we would rather say ourselves: we do not yet apply a cryptographic signature to every AI image. We have passed the technical review in the C2PA Conformance Program; inclusion in the Conforming Products List is still pending. As long as the signature is not active, it stays out of our promises. What we do check are provenance credentials in other people's images that you bring along.