EU AI Act: AI content labelling

Why this matters

From 2 August 2026, the transparency rules of the EU AI Act (Article 50) apply to organisations that publish digital content. In short: where AI has generated or altered content in a way that changes how real it appears to your audience, disclosure may be required. As the publisher, the disclosure obligation sits with you — and whether any specific asset needs a label is your team's decision, based on your own interpretation of the Act.

Papirfly supports different ways of labeling your AI Content. We preserve any AI-provenance information carried on your assets as they move through the platform, and we give you the tools to record, find, and disclose AI involvement in the way that fits your organisation. This guide explains what's changed, what Papirfly now offers, and the practical steps we recommend.

Note: This article is for guidance only and is not legal advice. For how the Act applies to your specific content, consult your own legal or compliance team.

What the Act actually asks for

The Act is concerned with content that could change how real something appears to your audience. In practice that means things like:

• AI-generated images or visuals of people, places, or events that look real
• AI-generated text published to inform the public without human editorial review

Not every AI-assisted edit needs a label. Routine adjustments — colour correction, lighting, noise reduction, cropping, file compression, accessibility improvements, minor cosmetic touch-ups — are generally outside the scope when they don't change how authentic the content appears. The line is drawn where AI changes the substance of what the viewer sees: inserting, removing, or replacing objects or people, face swaps, or generating realistic scenes that never happened.

Where a label is needed, it belongs on the content you publish — the website visual, the brochure, the social post — not buried in your asset library.

How Papirfly helps

We've added AI-provenance support across the platform so you can run your own process end to end.

Flag AI content on your assets. A standard AI field lets you mark each asset as AI Generated, AI Modified, Unknown, or No AI used, with a clear visual badge shown on AI content across the platform.

Find it easily. Filter your library by the AI field to locate AI-involved assets for review or audit — no manual hunting.

Bring your own data. Map your existing metadata into the AI field so assets are flagged automatically rather than by hand. You can also set the value manually (useful for tools like Photoshop that leave no trace), including in bulk.

Disclose your outputs. Add an AI label or disclaimer to documents produced in the platform — in your own wording and design, placed where and when you choose. The official EU AI icons are available, and where no AI content is detected, the labelling controls simply don't appear.

Preserve provenance. Where assets carry industry-standard content credentials (C2PA), we keep that information intact as they move through the platform, so the provenance chain your compliance process depends on isn't broken.

Transparency with our own AI tooling. Anywhere Papirfly uses AI, it's marked with an AI icon, so your team always knows when they're working with an AI feature.

Read more about how to enable this in place 

What we recommend as a minimum

Even if your approach is light-touch, three steps put you in a solid position:

1. Run a quick AI inventory. Note which tools and which assets in your library involve AI. This is the single most useful step before the deadline.
2. Start flagging AI content. Turn on the AI field and begin marking new AI-generated or AI-modified assets so provenance is captured from now on.
3. Decide where labelling lands. With your legal or compliance team, agree which of your published outputs need a visible label — and use Produce to add it.

Clients who want to go further can layer on automated metadata mapping, mandatory-on-upload rules, approval workflows, and template-level disclosure.

Frequently asked questions

1. Does this mean Papirfly makes us "AI Act compliant"? No - and you should be cautious of any vendor that claims to. Compliance is your organisation's responsibility, and whether a specific asset needs a label is your legal team's call. What Papirfly does is preserve AI provenance through the platform and give you the tools to record, find, and disclose AI involvement so you can run your own compliance process.

2. Do I have to label every image that touched an AI tool? No. The Act targets content where AI has generated or substantially altered what the viewer sees. Routine edits — colour and lighting correction, noise reduction, cropping, compression, accessibility tweaks — generally don't require a label. A useful test for product imagery: if the image still represents the product accurately, it's usually fine; if AI makes it look materially different, better, or higher-quality than reality, treat it with more care.

3. What happens if we don't do anything? Nothing changes automatically in your day-to-day — the platform won't block your work, and if you don't use AI-generated imagery, the labelling controls never even appear. But the disclosure obligation still applies to your published content from 2 August. We'd recommend at least running an AI inventory and turning on the AI field so you're capturing provenance going forward, even if you configure more later.

4. What is C2PA, and do we need it? C2PA (Content Credentials) is an industry standard that attaches verifiable origin and edit-history information to an asset — which camera or AI tool made it, what edits happened, and so on. You don't need C2PA to use Papirfly's AI field; it's one optional input. Where your assets already carry C2PA credentials, we preserve them through the platform so the information isn't stripped during conversion or export.

5. Can we use our own labels and wording instead of the EU icons? Yes. The official EU AI icons are available out of the box, but the Act allows an equivalent icon or label as long as it clearly carries "AI" as its main visual element. In Produce you can apply your own wording, styling, and placement to match your brand. The customisation in advanced templates is bespoke work — your CSM can scope that with you.

6. We're on Keepeek — does this work for us too? Yes, though the setup differs. Keepeek's flexible data model supports AI-provenance fields per client, mapped in from asset metadata. One thing to know: provenance doesn't automatically propagate into produced documents for Keepflying customers out of the box, so talk to your CSM about how to handle disclosure on final outputs.

7. Where does the label actually need to appear — in the library or on the published file? On what you publish. The obligation is about what your audience sees, so the label belongs on the final output — the exported document, the campaign visual, the social post. The AI field and badge inside Papirfly help you track and find AI content; Produce is where you apply the disclosure to the output itself.