Artificial intelligence has become part of everyday business processes. Companies use AI to communicate with customers, create content, run marketing, analyse data, automate internal workflows and develop new products.
But with the technology comes a new question for business: must a customer, employee or other user be told that they are interacting with AI, or that content was created or altered by artificial intelligence? Article 50 of the EU Artificial Intelligence Act (AI Act) answers this question to a large extent.
Since 2 August 2026, its transparency requirements have been legally binding. The European Commission has also published separate Guidelines explaining how these provisions apply in practice.
It is important not to read Article 50 as a simple rule that “everything created by AI must be labelled”. In fact, the law sets different obligations for different situations and for different players in the AI chain.
Who Article 50 applies to
The AI Act is built on a risk-based approach. Depending on the nature and level of risk, AI systems fall under different regimes, from a ban on certain practices to minimal requirements.
Article 50 is aimed mainly at systems where the legislator sees a risk that a person will not realise they are dealing with AI or with artificially created or altered content. These include systems that:
- interact directly with individuals;
- generate or manipulate audio, images, video or text;
- can produce deepfake content;
- are used for emotion recognition or biometric categorisation.
In other words, Article 50 is not only about developers of large AI models. It can also affect ordinary businesses if they use such an AI system in their operations.
Provider and deployer: which role does your company play
One of the key concepts of the AI Act is the distinction between provider and deployer.
- A provider is a person that develops an AI system, or has it developed, and places it on the market or puts it into service under its own name or trademark.
- A deployer is a person that uses an AI system under its own authority in a professional activity.
In practice, the same technology can create different legal obligations depending on who develops it, who places it on the market and who uses it. For example, a technology company that built its own generative AI system and offers it to clients may be a provider. A company that uses a ready-made AI service to work with customers or create content may, in the right circumstances, be a deployer.
So the first question for a business should not be “do we use AI?” but “what role do we play in relation to this AI system?”. The answer determines which requirements apply.
Three categories of transparency requirements
Article 50 does not set a single universal labelling rule. It provides for several separate categories of transparency.
1. People must know when they are talking to AI
If a system is intended to interact directly with individuals, it must be designed so that people are told they are interacting with AI, unless this is obvious from the context. This rule may apply to:
- chatbots;
- AI assistants;
- virtual agents;
- AI customer support systems;
- other tools that may give the impression of talking to a real person.
For business, this means AI in customer support is no longer only a matter of technology and customer convenience. You also need to think about how to tell customers who, or what, they are dealing with.
2. AI-generated content must be technically identifiable
A separate obligation applies to providers of systems that generate synthetic audio, image, video or text content. Such content must carry machine-readable marking that makes it possible to detect that it was artificially generated or manipulated.
This is fundamentally different from an ordinary “AI-generated” caption. Machine-readable marking is not meant to be seen by the consumer. It lets technical systems determine where the content came from. For this reason, companies that develop their own AI products need to address transparency at the level of system design and architecture, not after the product reaches the market.
3. Emotion recognition and biometric categorisation
Article 50 also sets requirements for operators of emotion recognition and biometric categorisation systems: people exposed to such a system must be informed that it is being used.
There is an important detail here: transparency cannot be considered separately from the AI Act’s prohibitions. For some uses of AI, the problem is not how to tell people about the system but that the practice may be banned outright. For example, the AI Act restricts emotion recognition based on biometric data in the workplace and in educational institutions.
So compliance in this area should begin by checking whether the use of the technology is permitted at all, and only then move on to the information requirements.
The extraterritorial reach of the AI Act
The AI Act has an extraterritorial element. A Ukrainian, British or US company should therefore not automatically assume that European regulation does not concern it. This matters most for businesses that:
- sell goods or services to customers in the EU;
- use AI systems in European operations;
- have users or customers in the EU;
- distribute AI-generated content to a European audience;
- operate through international digital platforms.
The geography of AI compliance should therefore be determined not only by where a company is registered. Look also at the system, how it is used, who the users are and its actual effect on the European market.
What changes for business
Article 50 changes the approach to using AI. Before, a company could ask only: “does this AI tool let us create the content we need?”. That is no longer enough. You also need to establish:
- who is the provider and who is the deployer;
- which AI tool is being used;
- what content it generates or changes;
- who will see that content;
- whether a person could take it for genuine;
- whether the use of AI must be disclosed;
- whether technical marking is in place;
- who is responsible for compliance.
And this is only the first level.
What comes next
In the next part, we will look at the question that is often the most practical for business: when AI content must be labelled for people, what counts as a deepfake under the AI Act, whether AI advertising must be labelled, and what to do with images, video, voices and AI-generated text.
Using AI in your business?
It is not enough for a company to know that the AI Act is in force. It needs to understand which provisions apply to a specific AI tool and a specific business process.
Our team helps businesses assess the legal status of AI solutions, determine provider and deployer obligations, and build practical compliance procedures in line with EU requirements. If your company already uses AI or is only planning to integrate it into its processes, get in touch. We will help you identify the legal requirements and risks before they turn into a problem for your business.