A new obligation entered into application on August 2, 2026
On August 2, 2026, a new block of obligations under the European AI Act became applicable: the transparency rules of Article 50. Unlike most of the regulation, which primarily targets "high-risk" AI systems (recruitment, credit scoring, healthcare...), Article 50 covers a much more common and widespread situation: simply exposing a customer or visitor to an AI system without clearly telling them.
In practice, if your site has a chatbot, if you publish AI-generated visuals, or if you use text generation for public content, this obligation applies to you — whether you're a multinational or a five-person SME. Legal analyses published on the topic agree: there is no general exemption based on company size.
What Article 50 actually requires
Article 50 bundles several distinct obligations, three of which apply directly to an SME's website or app:
- Chatbots and conversational assistants. Anyone interacting with an AI system must be informed they are not talking to a human, from the start of the exchange. A welcome message such as "Hello, I'm the virtual assistant for [your company]" is generally sufficient, as long as context leaves no room for doubt.
- AI-generated or manipulated content (deepfakes, visuals, text). Publishing AI content depicting real people, places, or events — especially on topics of public interest — must be disclosed as such.
- Emotion recognition or biometric categorisation. If a tool analyses emotions or categorises visitors based on biometric data, that use must be disclosed.
One technical point has been deferred: machine-readable marking (Article 50, paragraph 2 — automated, standardised tagging of AI content) benefits from a grace period until December 2, 2026 for systems already on the market, as detailed in the official AI Act guide to these transparency rules. But the disclosure obligation itself — the visible, human-readable message within the interaction — has applied without delay since August 2. This is the point many businesses get wrong, assuming "everything is pushed back to December."
Is your SME affected? Three common cases
| Situation | Obligation | What's usually enough in practice |
|---|---|---|
| Chatbot or conversational form on the site | Disclose from the start of the exchange | A clear welcome line, visible in the interface, not buried in the terms |
| AI-generated visuals or text published online (especially on current events) | Disclose that the content is AI-generated or manipulated | A discreet but readable notice near the content |
| AI-automated customer service (email, social media) | Same logic as chatbots: state it's an automated system | A signature or standard message from the first contact |
Many SMEs discover, once they look into it, that they tick at least one of these three boxes without ever having thought about it — a chatbot installed quickly via a plugin, or landing-page visuals generated by AI with no mention at all.
What to do right now
- Take stock of your public-facing AI tools: chatbot, image generator, automated social media replies, summaries or articles written with AI assistance and published under your name.
- Add a visible notice, not a buried one. Emerging guidance on this topic is clear: a line in the legal notices is not enough — the information must be perceivable at the moment of the interaction itself.
- Document the process. Even without a formal record-keeping obligation, being able to show you audited your AI usage and put the necessary notices in place is your best protection in case of a check.
- Review content already published, not just new content. The obligation concerns the current exposure of visitors, not only future publications.
- Train whoever manages your site or social media to build this reflex into future projects — chatbots, campaigns, redesigns.
What not to do
- Don't assume your company size protects you. There is no revenue threshold below which the disclosure obligation disappears — only the way the fine is calculated is proportionate.
- Don't confuse the December deadline with a general grace period. Only the automated technical marking is deferred to December 2, 2026; user disclosure already applies.
- Don't settle for a mention buried in the terms of use. That's not what the text requires: the information must appear within the interaction itself.
- Don't wait for an inspection to act. Bringing a chatbot or a content notice into compliance usually takes a few hours, not weeks.
The penalties at stake
Non-compliance with Article 50 falls under the sanctions regime of Article 99 of the AI Act, as summarised in coverage of the topic: up to €15 million or 3% of global annual turnover, with the higher amount applied to large companies. For SMEs, it works the other way around: the lower of the two caps applies, which mechanically limits financial exposure without removing the risk entirely — being publicly called out for a chatbot pretending to be human is never good publicity.
The link to your website and digital tools
This obligation touches directly the building blocks an SME puts in place to capture leads online: a lead-qualification chatbot, a visual generator for a campaign, blog content written with AI assistance. The good news is that compliance is rarely a heavy project — in most cases it's about adding a clear notice at one specific point in the interface, not rebuilding the tool.
At ÜMAIN, this kind of review and adjustment usually fits within our website maintenance offers, starting at €150 per month — a good opportunity, if you've never done this audit, to also check that your site is still performing well technically. If you're preparing a new site or a redesign, it's also the right time to build these notices in from the start rather than bolting them on afterwards — a topic we cover in our article on the signals that show a redesign is needed.
A useful side effect: more clarity, more trust
Beyond the legal obligation, this compliance work has a side effect rarely mentioned: a visitor who clearly knows they're talking to an automated assistant, or that a visual was AI-generated, tends to trust the rest of the site more. Conversely, discovering after the fact that you were misled on this point durably damages the relationship with a brand — especially for an SME whose local or sector reputation carries more weight than it would for an anonymous large group. Treating Article 50 as a transparency exercise rather than an administrative burden usually makes it a faster project, and sometimes even a differentiator against competitors who haven't thought about it yet.
In summary
Since August 2, 2026, Article 50 of the AI Act requires clearly informing any visitor who interacts with a chatbot or is exposed to AI-generated content — with no size exemption for SMEs. Only the automated technical marking is deferred to December 2026; the disclosure obligation itself already applies. Compliance is usually quick: take stock of your public-facing AI tools, add a visible notice in the right place, and document the process.
Not sure whether your site or chatbot is affected? Describe your situation in a few minutes and we'll get back to you with a reasoned answer within 24 hours. Describe your project.