Updated 1 October 2026 to reflect the Digital Omnibus, in force since 27 July 2026.
Since 2 August 2026, the EU AI Act asks something of almost every business that puts AI in front of people. It is not the heavy high-risk regime. It is a lighter duty that reaches far more companies: tell people when AI is involved.
While the high-risk obligations were deferred to 2 December 2027, the transparency rules in Article 50 were not. They have applied since 2 August 2026, and they catch almost any organisation running a customer-facing chatbot or generating content with AI. One date is still ahead: 2 December 2026, when the grace period for machine-readable marking ends. If you were waiting for the EU AI Act to affect you, this is usually where it does, first.
This guide covers what Article 50 requires, the watermarking grace period that softens part of it, who is on the hook, and what it pointedly does not demand. For the wider timeline, see the EU AI Act explained.
What Article 50 actually requires
Article 50 sets transparency duties for AI systems that interact with people or produce content. Four obligations matter in practice.
AI interaction disclosure. Systems that interact directly with people, such as chatbots and voice agents, must make clear that users are dealing with AI, unless that is already obvious from the context.
Synthetic content labelling. Providers of systems that generate or manipulate audio, image, video, or text must mark the output, in a machine-readable way, as artificially generated.
Deepfake and content disclosure. Deployers that publish deepfakes, or AI-generated text on matters of public interest, must disclose that the content is artificial.
Emotion recognition notification. Where emotion recognition or biometric categorisation is permitted at all, the people subject to it must be informed.
Grasp helps you find and track the AI tools in use, so the systems these duties touch are on your list rather than discovered late.
The watermarking grace period
One part of Article 50 still has runway. Under the Digital Omnibus (Regulation (EU) 2026/1744), providers of generative AI systems that were already on the market before 2 August 2026 have until 2 December 2026 to meet the machine-readable marking obligation in Article 50(2). Systems placed on the market since then must mark their output from the start, and the disclosure duties have applied to everyone since 2 August 2026.
Who Article 50 applies to
Both providers and deployers. Providers build the marking and disclosure capability into their systems, and deployers switch it on and disclose in their own use. The reach is wide: any organisation with a customer chatbot, an AI voice line, or a marketing function generating synthetic images, audio, or video is likely in scope. Like the rest of the Act, it applies extraterritorially, so non-EU companies serving EU users are included.
What Article 50 does not require
This is the part worth underlining, because it prevents over-spend. Article 50 is a transparency duty, not the high-risk regime. A chatbot that owes an AI disclosure does not, by virtue of Article 50, need a conformity assessment, a technical file, or EU database registration. Those belong to the high-risk tier, and only when a system is used in a high-risk context. Confusing the two is a common way teams inflate a transparency obligation into an imagined high-risk programme. The risk classification guide shows where each system actually lands.
How to comply
Three steps cover most of it. Inventory every user-facing AI system and every AI content workflow you run. Add a clear disclosure wherever AI interacts with people, and labelling or watermarking wherever AI generates content. Then document what you did, so you can show it on request. The compliance checklist folds this into the wider programme.
Why this is the obligation to act on first
Transparency has applied since 2 August 2026, the marking grace period ends on 2 December 2026, and standalone high-risk obligations follow on 2 December 2027. For most organisations that order is a gift: the obligation that landed first is also the lighter one. If your disclosures are not in place yet, treat Article 50 as the task for now and the high-risk file as the longer build, and you are sequencing the work the way the timeline actually runs.
Frequently asked questions
When do Article 50 obligations apply?
They have applied since 2 August 2026. Providers of generative AI systems already on the market before that date have until 2 December 2026 to meet the machine-readable marking duty under Article 50(2); the disclosure obligations apply to everyone now.
Does my chatbot need an AI disclosure?
Yes, if it interacts directly with people and it is not already obvious that they are dealing with AI. The disclosure is the core Article 50 duty for chatbots and voice agents, and it does not turn the chatbot into a high-risk system.
Do I have to label AI-generated content?
Yes. Providers must mark AI-generated or manipulated audio, image, video, and text in a machine-readable way, and deployers must disclose deepfakes and AI-generated content on matters of public interest.
What is the watermarking grace period?
Providers of generative AI systems placed on the market before 2 August 2026 have until 2 December 2026 to meet Article 50(2)'s machine-readable marking obligation, four months after the main Article 50 date. It does not cover newer systems or the disclosure duties.
Does Article 50 apply to companies outside the EU?
Yes. If your AI system interacts with, or produces content for, people in the EU, Article 50 applies regardless of where your company is based.
Grasp finds the AI tools in use across your organisation and keeps an owner and a decision on record for each, the inventory an Article 50 review starts from. See how Grasp finds your AI