The Digital Omnibus headline has been the postponement: high risk moves to December 2027, regulated products to August 2028. That's good news, and it gives real breathing room. But it has an uncomfortable side effect: it has drawn attention away from the obligation that does arrive this summer, and that affects far more companies than high risk does. We're talking about Article 50, transparency.
What Article 50 is, and who it reaches
Article 50 doesn't regulate "dangerous" systems. It regulates systems that, by their nature, can confuse a person about what or whom they are interacting with. And that category is extremely broad: it includes tools that any SME uses on a daily basis.
Article 50 has four paragraphs, each with its own subject:
- Interaction with AI (Article 50(1) — provider). When a person interacts with an AI system —the typical case is a chatbot—, they must know it, unless this is obvious from the context. The notice "you're talking to a virtual assistant" stops being good practice and becomes an obligation.
- Synthetic content (Article 50(2) — provider). AI-generated audio, image, video or text content must be marked in a machine-readable format and be detectable as artificially generated or manipulated.
- Emotion recognition and biometric categorisation (Article 50(3) — deployer). Whoever uses such a system must inform the natural persons exposed to it of its operation — the one paragraph that reaches an SME directly, if it uses these tools.
- Deepfakes and text of public interest (Article 50(4) — deployer). Content that imitates real people, objects or events and could appear authentic must be disclosed as artificially generated or manipulated; AI-generated text published to inform the public on matters of public interest must be declared as such, with some nuances.
The date, and the nuance the Omnibus does introduce
The general rule doesn't change: Article 50 applies from 2 August 2026, and it reaches both the systems already in use and those added later — the Regulation provides no transitional period at all for paragraphs 1, 3 and 4. The Omnibus does not postpone these transparency obligations.
The only temporal nuance affects the watermarking of synthetic content. The new Article 111(4) gives providers of systems already on the market before 2 August 2026 until 2 December 2026 to comply with Article 50(2). It's a period the Omnibus introduces, not one it shortens: systems placed on the market from 2 August onward comply from that date, with no transitional period.
That distinction between a "new system" and a "system already on the market" operates only for the marking under Article 50(2), which belongs to the provider. Even so, it makes something that sounds bureaucratic, but isn't, genuinely relevant: knowing which systems you already had running and which you've added since, because your inventory map decides what you have to verify, and from when.
Where this shows up in a real SME
You don't need to be a tech company to fall under Article 50. Three common examples:
The customer service or booking chatbot. A clinic that handles first appointments over WhatsApp, a distributor with automated customer support. If there's conversational AI behind it, the user must know they aren't talking to a person. It's the most common Article 50(1) trigger, and the easiest to resolve: a clear notice at the start of the conversation.
AI-generated marketing content. Images, videos or text produced with generative tools for social media or the web. This falls under synthetic content and, if it imitates real situations in a way that could appear authentic, under deepfakes too.
The copilot that drafts customer-facing text. When the generated text is published to inform the public on matters of interest, it must be declared. In corporate communications, it's worth having clear criteria for when this applies.
What to do before August
Article 50 transparency is, technically, one of the more manageable obligations under the Regulation. It doesn't require conformity assessments or technical documentation files. It requires judgement and execution:
- Identify in your inventory which systems interact with people or generate content.
- Determine which transparency obligation applies to each one — and, only for the marking under Article 50(2), whether the system was already on the market before 2 August 2026.
- Implement the notices and the marking: the chatbot disclaimer text, the synthetic content label, the internal policy on what gets declared and how.
- Document the decision. Doing it isn't enough; you should be able to show that you thought it through, and why you classified each system the way you did.
The postponement of high risk is real, and welcome. But it shouldn't obscure the obvious: 2 August 2026 is still a date with consequences. The company that uses the summer to close out its transparency work arrives without surprises. The one that assumes "everything got postponed" is in for one.
This article is for informational purposes only and does not constitute legal advice.