AI Act: Transparency Obligations

You boot up your professional laptop and are greeted by your company’s AI agent. You interact with the agent on a daily basis; you ask it for information and to execute tasks on your behalf. At times, you use the images and text it generates for LinkedIn posts. Some of its content is posted on the company’s website as well.  You are aware that this is all AI, but are unsure which rules apply to it. Welcome to our five-part series on transparency obligations for certain AI systems.

Our first instalment explores the scope of Article 50 of the EU AI Act. Each subsequent article focuses on a specific transparency scenario, using common examples to explain what the rules mean in practice and how organisations can meet their compliance obligations.

In part two, we talk about the rules applicable to chatbots and other interactive systems. Part three focuses on the obligation to mark AI-generated content in a machine-readable format. The fourth instalment addresses organisations which use AI systems capable of recognising emotions or categorising individuals based on biometric data. In the final part, we discuss the rules on deepfakes and AI-publications aimed at informing the public on matters of public interest.

1. Introduction

Reflecting its dual focus on product safety and fundamental rights protection, the EU Artificial Intelligence Act (AI Act) places significant emphasis on transparency. Recalling the 2019 ‘Ethics Guidelines for Trustworthy AI’, the EU legislator describes transparency in recital 27 of the AI Act as a requirement which ensures that “AI systems are developed and used in a way that allows appropriate traceability and explainability, while making humans aware that they communicate or interact with an AI system, as well as duly informing deployers of the capabilities and limitations of that AI system and affected persons about their rights”.

Article 50 of the AI Act is arguably the main transparency provision of the Act. It targets situations in which people may be misled about whether they are interacting with AI or whether content has been generated or manipulated by AI, and establishes a specific set of transparency obligations for providers and deployers of AI systems that carry such risks.

In an attempt to provide guidance regarding these obligations, the European Commission has recently published dedicated guidelines on Article 50 (EU Guidelines). Furthermore, companies have the option to sign the ‘Code of Practice on Transparency of AI-Generated Content’ (EU Code of Practice), adherence to which demonstrates compliance with the marking and labelling requirements of Article 50. These materials provide considerably more practical detail than the wording of Article 50 itself and should form part of any compliance assessment.

2. Scope of application

2.1 Who is covered?

The starting point is to determine whether the organisation is a ‘provider’ or a ‘deployer’ under the AI Act, since Article 50 allocates different obligations to providers and deployers.

A provider is an entity that develops an AI system, or has an AI system developed, and places it on the EU market or puts it into service under its own name or trademark. Importantly, providers do not have to be established in the EU to fall within the AI Act. The Regulation also applies to certain providers and deployers established outside the EU where the output produced by their AI system is used in the Union.

         The company who made the chatbot is the provider.

A deployer is an entity which uses an AI system under its own authority for professional purposes; where the system is used for purely personal, non-professional activities, the user is not considered a deployer.

The company who uses the chatbot is the deployer. Provider and deployer can be the same company!

An employee using an AI system on behalf of an organisation is generally not treated as a separate deployer; the organisation remains the relevant deployer. The same applies where contractors or freelancers operate the AI system on the organisation’s behalf and under its responsibility and control.

It is important to note that each transparency requirement is subject to exceptions for law enforcement, which will not be discussed hereafter.

2.2 The four transparency  scenarios

Article 50 can broadly be divided into four scenarios – one per paragraph –  that trigger transparency obligations:

• 50(§1) – the AI system is intended to interact directly with individuals;

The classic example of a chatbot, both internally and externally.

• 50(§2) – the AI system generates synthetic content;

Image or video generation.

• 50(§3) – the AI system is an emotion recognition system or a biometric categorisation system;

A chatbot which also adapts its tone of message to your emotions.

• 50(§4) – the AI system generates or manipulates deep fake content or is used to generate or manipulate text published with the purpose of informing the public on public interest matters.

Cloning the voice of a person.

Article 50(§5) then imposes a horizontal requirement: information required under the four scenarios must be provided clearly and distinguishably, at the latest at the time of the first interaction or exposure, and must comply with applicable accessibility requirements.

2.3 Compliance deadlines

The transparency obligations generally took effect on 2 August 2026. There is, however, a specific transitional deadline for Article 50(§2): where an AI system generating synthetic audio, image, video or text content was already placed on the market before 2 August 2026, its provider has until 2 December 2026 to take the necessary steps to comply with the marking and detectability requirements. This is the only Article 50 obligation for which the AI Omnibus introduced such a grace period. Content generated before 2 August 2026 does not have to be labelled retrospectively.

In the next instalment of the series, we discuss Article 50(§1) in depth, and provide steps to ensure compliance when putting an interactive AI system on the market.