Since 2 August 2026, the transparency rules of the European AI Act have been enforceable. Plenty has been said about delays over the past few months, and that is fair — a good chunk of the regulation has indeed been pushed back. Just not the part most likely to touch you as an organiser. So the date is no longer an announcement: anything you have not arranged is now overdue.
Put simply: if an AI is talking or writing on your behalf anywhere in your ticket sales or communication, that has to be recognisable. Below we walk through what applies, what does not, and what to arrange in practice. Spoiler: for most organisers this is an afternoon of work, not a legal project.
Updated on 17 August 2026: the passage on the Code of Practice has been corrected. The Commission found the code adequate on 8 July 2026 and the AI Board did the same a day later, which means signatories can in fact rely on it to demonstrate compliance. Dutch supervision has been corrected too: the data protection authority and the RDI share the coordinating role. Earlier addition (3 August 2026): Article 50 actually became applicable on 2 August 2026 — the text and the checklist are now in the present tense. Earlier addition (27 July 2026): the Digital Omnibus package was published as Regulation (EU) 2026/1744 (Official Journal, 24 July 2026) and entered into force on 27 July 2026; on 20 July 2026 the European Commission adopted official guidelines on Article 50.
What was delayed and what wasn't
The confusion comes from the Digital Omnibus package. The European Parliament adopted it on 16 June 2026, the Council gave its final green light on 29 June, and the act was signed on 8 July. On 24 July 2026 it was published in the Official Journal as Regulation (EU) 2026/1744, and it entered into force on 27 July 2026. The package moves the obligations for high-risk AI systems to 2 December 2027 (stand-alone systems) and 2 August 2028 (systems embedded in products).
That deferred category covers things like AI in recruitment, credit scoring or critical infrastructure. Ticketing and event communication are not on that list — and the transparency obligations in Article 50 were not postponed. They took effect on 2 August 2026.
One detail: for AI systems already on the market before 2 August, the machine-readable marking of generated content has a run-off until 2 December 2026. That duty sits with the builders of those systems, not with you as a user — a system placed on the market after 2 August has to comply straight away.
What Article 50 says in plain language
Article 50 of the AI Act splits its duties across two roles: the provider (whoever builds or places the AI system on the market) and the deployer (whoever puts it to use). As an organiser you are almost always the second. That helps: the heavy lifting sits with your supplier.
The core of the article, translated to an event setting:
1. If an AI talks to your visitor, the visitor must know. A chatbot on your event page, an AI assistant answering questions about opening times or tickets — the visitor has to realise there is no human on the other end. Unless that is already obvious to a reasonably observant person.
2. Generated image, audio and video must be marked machine-readably. This is a duty for the provider of the AI tool, not for you. Use a generator that handles it properly and you are covered.
3. If you deploy emotion recognition or biometric categorisation, you must inform people. Think of cameras at the gate measuring "crowd mood". Alongside the AI Act, the GDPR applies in full here — and note that some applications are outright prohibited under Article 5 of the AI Act. More on the data protection side in our piece on what your ticket platform does with visitor data.
4. Deepfake-style imagery must be labelled. If you publish image or video content that is artificially generated or manipulated and resembles real people, places or events, you have to disclose that it is not real. For artistic or creative work a lighter form of disclosure is allowed, so your aftermovie effects survive intact.
5. For text the duty is far narrower than people assume. The labelling duty for AI-written text covers publications informing the public on matters of public interest, and falls away once a human has editorially reviewed the text and takes responsibility for it. An AI-drafted event description that you read through before publishing does not qualify.
The rule behind all the rules: clear, on time, accessible
What often gets skipped is the closing paragraph of Article 50. All information you must give under this article has to be clear and distinguishable, provided no later than the first interaction or exposure, and must meet accessibility requirements.
That last one is not a footnote. A grey 10-pixel line at the bottom of your chat window does not qualify, and neither does a label a screen reader cannot announce. You are immediately in the same territory as the European Accessibility Act, which requires your ticket shop to be accessible. If you have that in order already, there is little extra work here.
On 10 June 2026 the European Commission published a Code of Practice on marking and labelling AI-generated content. Signing remains voluntary, but the code has since become a recognised compliance instrument: on 8 July 2026 the Commission adopted an opinion finding that the code adequately covers the obligations in Article 50(2), (4) and (5), and the AI Board adopted its adequacy assessment a day later. The Commission now states that signatories "can rely on its measures to demonstrate compliance". If you do not sign, you have to show for yourself that your own measures are adequate. When enforcement started on 2 August 2026 the Commission published a first list of more than 180 signatories. In addition, on 20 July 2026 the Commission adopted official, non-binding guidelines on Article 50; they explain per role when the transparency duty applies and how to meet it — the most authoritative reference available right now.
Who supervises this in the Netherlands?
On 20 April 2026 the Dutch government took a step on supervision of the European AI rules. The principle is that existing regulators supervise AI within their own domain, so businesses "deal as much as possible with bodies they already know". The Dutch data protection authority (AP) and the Dutch Authority for Digital Infrastructure (RDI) share the coordinating role; for the areas without an obvious existing regulator — the transparency obligations among them — the AP is the proposed supervisor, with a dedicated AI board member. The bill that settles this definitively was open for consultation until 1 June 2026 and is still in progress.
For readers outside the Netherlands. The AI Act itself is an EU regulation and applies identically across all member states — the dates, Article 50 duties and fine tiers below are the same wherever you operate. What differs per country is which national authority enforces it. Check your own member state's designated market surveillance authority.
The fines sit in Article 99 of the regulation. A breach of Article 50 falls into the middle tier: up to 15 million euro or 3% of worldwide annual turnover, whichever is higher. For SMEs and start-ups the lower of those two applies instead. Those are numbers that sound unrealistic for a regional festival organiser, and they are — regulators rarely open with a maximum fine. But that is no reason to skip a label you can add in ten minutes.
A practical checklist: what to arrange now
- Map where AI talks or writes. Chatbot on your site, AI in your mail tool, an image generator for your promo, a smart search in your ticket shop.
- Check who the provider is per tool. If a product goes to market under your own name and branding, you can be treated as the provider yourself — ask your supplier how that is set out contractually.
- Label your chatbot visibly, in the first message or the header of the chat window. Not tucked away on a terms page.
- Check that the label is accessible: sufficient contrast, readable text size, and announceable to screen readers.
- Add a clear notice to AI-generated imagery that looks real. An illustration nobody mistakes for a photo is a different case from a "photo" of last year's edition that was never taken.
- Record who reviews your AI-drafted text. Human editorial review is good for quality anyway, and it also takes you out of the labelling duty.
How we handle it at MijnEvent
We did not have to change much, because this was the starting point already. Minty, our AI assistant for visitors and organisers, introduces itself as an AI assistant in its first message — not as a staff member with an invented name. The label sits in the chat window itself, at the same contrast and text size as the rest of the interface.
The AI helpers in the dashboard deliberately produce drafts only. An AI-generated event description or confirmation email arrives as a draft you review and publish or send yourself. That is not just nicer to work with — it is exactly the human editorial step Article 50 points to.
And we do not do emotion recognition or biometric categorisation of visitors. Our scanner at the gate reads a QR code and nothing else: no faces, no mood measurement, no profiles.
Want to see how that works in practice? Our pricing page sets out exactly what each plan costs and what is included — a flat fee per ticket, no percentage of your revenue. You can register your organisation for free and try everything in demo mode before selling a single ticket.
