The EU AI Act has been in force since August 2024 and is being phased in gradually. The next phase is a significant one: from 2 August 2026, the transparency obligations set out in Article 50 will apply to providers and operators of AI systems – including companies that merely use tools such as ChatGPT or Claude. At the same time, the EU has pushed back several other deadlines with the Digital Omnibus in June 2026. The deadline for labelling remains unaffected by this, which is precisely why it is worth taking a look at the details now.
What Article 50 requires from 2 August 2026
Article 50 covers situations where people interact with AI without necessarily realising it. Anyone using a chatbot in customer service or on their website must, in future, inform users that they are communicating with a machine. This information must be provided no later than the first interaction, in a clearly visible manner and in understandable language. An exception applies only if the use of AI is obvious to an averagely informed user.
The second major area of focus is the labelling requirement for synthetic content. Providers of systems that generate audio, images, videos or text must label the results as AI-generated in a machine-readable format, for example via watermarking such as Google’s SynthID or via embedded C2PA metadata, as used by OpenAI in generated images. Operators of deepfakes must also clearly disclose that the content has been artificially generated or manipulated. You can find the full text of the regulation, including all exemptions, on EUR-Lex.
Who is affected by the obligations and what penalties apply in the event of non-compliance
The AI Act distinguishes between two roles: providers develop and distribute AI systems, whilst operators use them in a professional capacity. The transparency obligations apply to both, which is why small and medium-sized enterprises must also address this issue, as the Chamber of Industry and Commerce (IHK) emphasises. Anyone using generative AI for product images, short video clips or entire campaigns is legally classified as an operator and has specific obligations. In the case of AI-generated texts that inform the public about matters of public interest, the regulation requires disclosure, unless a human can be shown to bear editorial responsibility.
Brussels is cracking down hard on sanctions: breaches of Article 50 can be punished with fines of up to 15 million euros or 3 per cent of global annual turnover, whichever is higher. That’s a hefty sum. For your day-to-day work with AI in marketing, this means: documenting processes, defining responsibilities and factoring in labelling requirements right from the start. If you set this up properly, you’ll turn this obligation into a sign of trust for customers and business partners.
The Digital Omnibus postpones other deadlines
The EU’s Omnibus Package, which was adopted in June 2026 and reschedules several elements of the AI Act, has recently caused further confusion. The European Parliament gave its approval on 16 June, the Council followed suit on 29 June, and the European Commission has published the details of the legal framework. The obligations for high-risk systems, in particular, have been postponed, whilst Article 50 will come into force as planned. The reasons behind the postponement are a lack of technical standards and concerns about the competitiveness of European companies, resulting in the following timetable for you:
- 2 August 2026: The transparency and labelling obligations under Article 50 will become binding on providers and operators.
- 2 December 2027: The obligations for autonomous high-risk systems under Annex III come into force; this was originally scheduled for 2 August 2026.
- 2 August 2028: AI in regulated products listed in Annex I will be covered in the final phase; the previous deadline was August 2027.
How to get your business ready in good time
Sounds like a lot of work? With a structured roadmap, you’ll get to grips with the requirements quickly, and much of it is already part of professional AI implementation anyway. The more autonomous the systems become, the more important the foundations become: An AI agent now handles entire workflows, and models such as Claude Fable 5 carry out tasks that used to require entire teams. If you’d like some support with this, you might know of a digital agency in Düsseldorf that’s quite well versed in AI – these five steps will get you ready by the deadline:
- Carry out an AI inventory: Identify which tools are in use within your organisation and whether you act as a provider or an operator in each case. This division of roles determines exactly which obligations apply to you.
- Updating chatbots: Add a clear notice stating that users are communicating with an AI, ideally right at the start of the conversation. A short sentence in the first dialogue window is sufficient to meet this requirement.
- Automate labelling: Check whether your generators apply machine-readable labels, and add visible warnings for deepfakes. For tools without built-in labelling, you should switch to alternatives.
- Train your team: The requirement for AI literacy is explicitly enshrined in the AI Act; regular training directly contributes to your compliance.
- Set up your documentation: Keep a record of who creates what content using which tool; this will make any subsequent audit much easier.
FAQs on the EU AI Act
What does Article 50 of the EU AI Act cover?
Article 50 sets out the transparency obligations under the EU AI Act. It requires providers and operators to disclose the use of AI, for example in chatbots, AI-generated content and deepfakes. The rules will be binding across the EU from 2 August 2026. The aim is to ensure that people can always recognise when they are interacting with AI or viewing AI-generated content.
Who will be affected by the transparency obligations from August 2026?
This applies to providers of AI systems as well as operators – that is, organisations that use AI in a professional capacity. This means that small and medium-sized enterprises are also subject to these obligations as soon as they operate chatbots or create and publish content using AI. Article 50 applies regardless of turnover or company size; the only exception is purely private use.
How must organisations label synthetic content?
AI-generated audio, image, video and text content must be labelled in a machine-readable format, for example using watermarks or metadata. In the case of deepfakes, a visible disclosure is also required so that users can immediately recognise the artificial origin. The technical labelling is the responsibility of the provider of the relevant AI system.
What penalties apply for breaches of the EU AI Act?
Breaches of the transparency obligations may result in fines of up to 15 million euros or 3 per cent of global annual turnover, whichever is higher. Article 99 of the Regulation applies. For small and medium-sized enterprises, the lower of the two amounts applies as the upper limit.
Does the labelling requirement also apply to marketing texts?
Visible disclosure is mandatory for texts, particularly where they inform the public about matters of public interest and editorial control lies solely with the AI. For advertising copy that is reviewed and approved by a human, machine-readable labelling by the AI tool provider is sufficient. You can still opt for voluntary labelling, however; many customers appreciate this openness.
