The EU AI Act and Your DAM: What Article 50 Means for New and Legacy Assets
- Article 50 applies from 2 August 2026; older content does not have to be labelled retrospectively.
- Reuse is where legacy assets create risk: you cannot disclose what you never recorded.
- An ‘AI involvement’ field, filled in by experts and checked by consensus, makes your answers defensible.
What does the EU AI Act say about AI-generated content?
Article 50 of the EU AI Actsets transparency rules for AI-generated and AI-manipulated content. It applies from 2 August 2026.
Providers
Providers of generative AI systems must mark outputs such as images, audio, video and text in a machine-readable way, so they can be detected as AI-generated.
Deployers
Deployers must disclose deepfakes, and AI-generated or manipulated text published to inform the public on matters of public interest. The Commission’s guidelines on Article 50 explain both duties in detail.
The Digital Omnibus left most of Article 50 on its original timeline. It added one grace period: generative systems already on the market before 2 August 2026 have until 2 December 2026 to meet the marking requirement.
In June 2026 the European Commission published a final, voluntary Code of Practice on marking and labelling AI-generated content. It suggests layered marking, such as metadata plus watermarks.
Does the EU AI Act apply to old assets in my DAM?
Not retrospectively. Content generated before 2 August 2026 does not have to be labelled after the fact, although the Commission encourages voluntary labelling.
That does not make the archive irrelevant. Assets are reused. When an old image goes into a new campaign, report or public statement, someone has to decide whether it needs a disclosure. That decision depends on knowing whether AI was involved, and most DAM records cannot tell you.
Which legacy assets are most likely to involve AI?
- Retouched product photography, where generative fill or AI tools were used in editing.
- Upscaled archive images, where software added pixels that did not exist in the original.
- Composite images, where one layer came from a generative tool and nobody logged it before the file was flattened.
- AI-written descriptions and captions added by auto-tagging tools.
If your metadata schema has no reliable field for AI involvement, you do not just have a gap. You have “compliance unknown”, which is harder to defend.
Why can’t detection tools answer the question?
Detection helps, but it struggles with older edits, partial manipulation and flattened composites. In the end someone has to look, confirm and record an answer that will stand up later. That is a review workload, and most teams cannot absorb it as a side project.
How do DAM teams build a defensible AI-involvement record?
- Add the field. Create an “AI involvement” field with clear values, such as none, AI-assisted edit, AI-generated element, fully AI-generated, unknown.
- Prioritise by reuse. Start with assets most likely to be published again: brand, product and campaign collections.
- Ask a yes-or-no question. Show each asset to people who know its history: was AI involved in creating or editing this?
- Require agreement. Accept the answer when reviewers agree; route disputes to the asset’s owner.
- Keep the trail. Record who confirmed each answer and when.
The Code of Practice is relevant here too. AI-generated text that has been through human review or editorial control, with someone taking editorial responsibility, is exempt from the labelling duty for public-interest text. Recorded expert review is exactly that kind of control.
Expert in the loop (EITL) is a review model in which the people who know the content best (brand managers, product specialists, archivists, rights owners) confirm, correct or reject metadata, including AI-generated tags, and their decisions are checked by peer consensus before they become the permanent record.
How expert-in-the-loop review works in TagQuest
How TagQuest helps
TagQuest runs this review as a Yes or No Quest. Invited experts confirm AI involvement one asset at a time. Consensus scoring turns a large backlog into a weekly routine, and points and recognition keep subject experts engaged.
Regulation will not wait for perfect automated detection. Teams with an expert-reviewed record of AI involvement will not be scrambling when the question lands on their desk.
This article is general information, not legal advice. Check your obligations with your legal team.
Frequently asked questions
When does EU AI Act Article 50 apply?
From 2 August 2026. Generative AI systems already on the market before that date have until 2 December 2026 to meet the Article 50(2) marking requirement.
Does the EU AI Act require labelling of old content in a DAM?
No. Content generated before 2 August 2026 does not have to be labelled retrospectively, although voluntary labelling is encouraged. Reusing that content may still raise disclosure questions.
How can DAM teams record whether AI was involved in an asset?
Add an AI-involvement field, ask experts who know each asset’s history to confirm it with a yes-or-no review, accept answers by consensus and keep a record of who confirmed them.
Who must comply with Article 50, providers or deployers?
Both, for different duties. Providers mark AI outputs in a machine-readable way; deployers disclose deepfakes and certain AI-generated public-interest text.