Your event app’s chatbot answered attendee questions all summer: travel routes, session times, badge pickup. Since 2 August 2026, it must also tell every attendee that it is a machine. That is not a courtesy. It is Article 50 of the EU AI Act, and it has been enforceable for two weeks.
The good news sits one layer deeper. Not every AI tool in your stack triggers a legal duty, and several duties that do exist sit with your software vendor rather than with you. The organisers who work through their stack once, scenario by scenario, are done in about half an hour. This guide shows you how.
The short version: since 2 August 2026, Article 50 of the EU AI Act (Regulation (EU) 2024/1689) requires that people are informed when they interact with AI systems such as chatbots, and that AI generated images, audio, video and certain texts are disclosed as artificial. The rules bind providers and professional deployers, which includes event organisers. Fines reach 15 million euros or 3 per cent of worldwide annual turnover.
What changed on 2 August 2026?
The AI Act entered into force back in 2024, but its transparency chapter only became applicable and enforceable on 2 August 2026. The European Commission adopted practical guidelines on 20 July 2026, and national market surveillance authorities can now investigate and fine. For organisers this ends a comfortable waiting period: the duties apply to AI systems already in use, not only to new purchases.
Article 50 at a glance
What non-compliance can cost since 2 August 2026
One widespread misunderstanding deserves clearing up first. In May 2026, the EU provisionally agreed to postpone the high risk obligations of the AI Act to December 2027. Several legal commentators publicly corrected their own earlier articles because they had assumed the whole package was delayed. It was not. The transparency duties in Article 50 arrived on schedule. If your managing director waves the topic away with “that AI law got postponed”, the answer is: the part that covers your chatbot and your AI visuals did not.
Two roles decide who owes what. A provider develops an AI system and places it on the market. A deployer uses an AI system professionally, under its own authority. An event organiser running a third party event app with an AI concierge is a deployer. That distinction runs through everything that follows.
The four scenarios that trigger duties
Article 50 does not regulate “AI in general”. It names four specific situations. Map your stack against them and most of the fog lifts.
| Scenario | Typical event example | Who is responsible |
|---|---|---|
| AI interacts directly with people | Chatbot or AI concierge in your event app or on the event website | Provider must design the system so people know they are talking to AI |
| AI generated image, audio, video or text | Key visuals, promo clips, AI voiceovers, auto generated session summaries | Provider must embed machine readable marking |
| Emotion recognition or biometric categorisation | Camera analytics measuring audience mood, badge cameras sorting attendees | Deployer must inform the people exposed |
| Deepfakes and AI text on matters of public interest | A realistic AI clip of your venue or a real speaker; AI written news style content | Deployer must disclose visibly, at first exposure |
Three practical consequences follow from the fine print of the Commission guidelines. First, purely assistive AI use, such as standard photo editing or grammar correction, does not trigger marking. Second, AI text that a human has editorially reviewed and takes responsibility for is exempt from the public interest text duty. Third, content that was both generated and published before 2 August 2026 does not need retroactive labels, but material generated earlier and published now does. That last point catches event marketers who sit on a folder of AI visuals from spring.
The scenario table also shows where your real exposure lies. For chatbot disclosure and machine readable marking, the legal duty sits with the provider. Your job is to buy from providers who have done it. For deepfake style content and emotion recognition, the duty sits with you as deployer, and the Commission FAQ is explicit that a visible label is required at first exposure. You cannot point to an invisible watermark embedded by the vendor.
Which duties sit with you, which with your vendor?
The provider and deployer split sounds academic until procurement season. Then it becomes the sharpest question you can ask.
If you publish an AI generated aerial clip of your venue that looks real, you owe the visible disclosure, not your design agency’s AI tool. A caption such as “AI generated visualisation” at first exposure does the job. The Commission published a voluntary Code of Practice in June 2026, together with a set of EU labelling icons, which is currently the closest thing to a template.
If your event app ships an AI assistant, the disclosure design duty belongs to the app provider. Your risk is choosing a vendor who has not built it in. So put three questions into every software evaluation: how does your chatbot identify itself as AI, how are generative outputs machine readably marked, and can you show documentation for both. A vendor who answers in writing removes your biggest Article 50 exposure in one email. We work through the same vendor logic for data protection in our GDPR compliant event software checklist, and the two checks combine well into a single procurement routine.
Solution: Converve approaches this from the transparent side of the spectrum. Matchmaking runs on a rule based meeting matrix that organisers configure and can explain, with an audit trail instead of a black box, and the B2B matchmaking platform documents how suggestions come about. Transparency duties are easier to meet with software that has nothing to hide.
What about AI matchmaking suggestions?
The question our industry is actually debating: if an algorithm suggests whom an attendee should meet, is that an Article 50 case? The Events Boardroom, one of the few industry voices to address the rules head on, argues that AI generated match suggestions need a disclosure too, and doubts that “many event organisers have clocked that it applies to them”.
The precise answer depends on mechanics. A rule based meeting matrix that applies criteria an organiser defined is not an AI system talking to attendees; the disclosure logic does not bite. A machine learning engine that chats with participants about who they should meet clearly is the chatbot scenario. Between those poles, disclosure is cheap insurance. A one line note in the app, “suggestions generated with AI support”, costs nothing, builds trust, and answers the question before a regulator or an exhibitor asks it. If you want the technical background on how matching engines differ, our guide on how AI can improve matchmaking accuracy walks through it.
Trust is the underrated half of this. Attendees who discover after the event that the “concierge” was a bot feel misled. The same disclosure that satisfies Brussels keeps your net promoter score intact.
Your event stack in 30 minutes
The compliance check is a walkthrough, not a project. Take the five steps in order.
The 30-minute check
Five steps through your event stack
- 1 Inventory List every AI touchpoint: app chatbot, AI visuals, promo clips, auto generated summaries, camera analytics, matchmaking engine, AI voiceovers.
- 2 Map scenarios Assign each touchpoint to one of the four Article 50 scenarios, or to none. Assistive editing and human reviewed text drop out here.
- 3 Split duties Mark each remaining item as provider duty or deployer duty. Your own to-do list is usually shorter than feared.
- 4 Ask vendors Send the three questions on chatbot self-identification, machine readable marking and documentation to every relevant vendor. Keep the written answers.
- 5 Set your house standard Define one labelling wording for AI generated content, brief marketing and agencies, and document the decision with a date.
Step five is what turns a one off check into a defensible position. When the managing director asks in November whether the winter congress is clean, the answer is a one page document with an inventory date, not a shrug. Document once, reuse every event.
Deadlines belong in the same document. Since 2 August 2026 the duties apply and are enforceable. On 2 December 2026 the transitional period ends for generative systems that were already on the market, which means legacy tools lose their grace on machine readable marking. On 2 December 2027 the postponed high risk obligations arrive, which matter to organisers mainly if AI screens applicants, including temporary event staff. Sustainability claims follow the same pattern of staged EU deadlines, and our guide to the EU Green Claims rules for event organisers covers the parallel track.
One more scope note that surprises many teams: like the General Data Protection Regulation (GDPR), the AI Act follows the market, not your registered office. A UK or US organiser whose event app serves attendees in the EU is in scope.
Frequently asked questions
Does my event chatbot have to say it is an AI?
Yes, unless it is already obvious to an averagely attentive person. The design duty sits with the chatbot provider, so verify it before you buy rather than after you launch.
Do I have to label every AI generated event image?
No. The duty covers content that could be mistaken for authentic, especially realistic depictions of people, places or events. Machine readable marking is the provider’s job; a visible label at first exposure is yours for deepfake style material. Abstract AI artwork and assistive edits are out of scope.
Do AI written invitation emails need a label?
Generally no. The text duty covers published text that informs the public on matters of public interest, and it falls away when a human reviews and takes editorial responsibility. Routine event copy with human review does not trigger it.
Do matchmaking suggestions need a disclosure?
A rule based meeting matrix does not fall under the chatbot scenario. AI driven suggestion engines that interact with attendees should disclose, and industry commentary already treats disclosure as expected practice. A one line note is cheap insurance either way.
What about content we generated before 2 August 2026?
If it was generated and published before that date, no retroactive labelling is required. If you generated it earlier but publish it now, the duties apply.
Who enforces this, and what can it cost?
National market surveillance authorities, in Germany for example the Bundesnetzagentur. Fines reach 15 million euros or 3 per cent of worldwide annual turnover, whichever is higher.
Conclusion: a half hour now beats a scramble in December
Article 50 rewards exactly the organisers who like clear processes. Four scenarios, a provider or deployer answer for each, three vendor questions, one documented house standard. The transitional deadline of 2 December 2026 gives you a natural date to have the walkthrough finished, and everything you document now carries over to every future event.
If you want event software where transparency is the starting point rather than a retrofit, with explainable matchmaking and documentation your legal team will actually enjoy reading, get in touch with Converve. We are happy to show you how the meeting matrix works, in plain language, from a human.
This article provides general information, not legal advice. For binding assessments of your specific setup, consult qualified counsel.
