Transparency Obligations Under the EU AI Act

See also our full guide to the EU AI Act for the wider context.

What are the transparency obligations?

The transparency obligations in Article 50 of the AI Act aren't about whether an AI system is allowed in the first place. They're about making sure people can know when they're dealing with AI, or when content has been created or manipulated by AI. The rules have applied since 2 August 2026 and were not affected by the later deferral of the high-risk deadlines.

The European Commission consulted on draft guidelines for Article 50 and approved the content of the final version on 20 July 2026, days before the obligations took effect. There is also a voluntary “Code of Practice on Transparency of AI-Generated Content”, published on 10 June 2026, which can help providers and deployers meet the labelling and disclosure requirements in practice. It isn't a legal requirement in itself. Under the guidelines, adhering to a code of practice that the Commission has assessed as adequate is one straightforward way to demonstrate compliance — though it grants no presumption of conformity, and compliance can equally be shown by other adequate means. Whether this particular code has been through that assessment isn't something our sources establish.

Who do the rules apply to?

Article 50 addresses both providers and deployers, but which obligation falls on whom depends on which part of the article is in play:

• Providers of AI systems that interact directly with people (chatbots, for example), and providers of AI systems that generate synthetic audio, image, video, or text content.
• Deployers of emotion recognition or biometric categorisation systems, and deployers of systems that create deepfakes or generate AI text published to inform the public on matters of public interest.

AI interactions must be recognisable

Where an AI system is intended to interact directly with people (typically a chatbot or similar), the provider must build it so the person is informed they're interacting with AI, unlessthat's obvious to a reasonably well-informed, observant and circumspect person in the circumstances. There's an exception for systems authorised by law to detect, prevent, investigate or prosecute criminal offences, unless the system is available for the public to report an offence.

The Commission's guidelines on Article 50 explain that AI agents fall under this obligation where they can interact with the people instructing them, or with other people while carrying out a task — making bookings, handling correspondence or completing purchases, for example. The guidelines go further than the Article itself here: they say an agent should disclose both that it is artificial and who it is acting on behalf of. That is the Commission's guidance rather than the wording of the Act, and the guidelines are not binding.

AI-generated content must be labellable

Providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text content must ensure the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated.

Three exceptions apply. The obligation doesn't apply to the extent the system performs an assistive function for standard editing, ordoesn't substantially alter the input data provided by the deployer or its semantics. Nor does it apply where the system is authorised by law to detect, prevent, investigate or prosecute criminal offences.

Systems already on the market before 2 August 2026 have until 2 December 2026 to meet this requirement.

The guidelines describe machine-readable marking as marks structured so that software can identify, recognise and extract them without a person involved — watermarks, metadata identifiers or cryptographic methods for proving provenance, for example.

Emotion recognition and biometric categorisation

Deployers using AI for emotion recognition or biometric categorisation must inform the people exposed to it that the system is in use, and handle the personal data in line with data protection law. There's an exception where the system is permitted by law to detect, prevent or investigate criminal offences. This sits alongside the separate, stricter restrictions that already apply to certain uses in workplace and education settings.

Deepfakes and AI-generated content of public interest

Deployers using AI to create or manipulate image, audio, or video content that constitutes a deepfake (content resembling existing people, objects, places, entities or events, which would falsely appear authentic or truthful) must disclose that the content has been artificially generated or manipulated. The same applies to AI-generated or manipulated text published to inform the public on matters of public interest, subject to certain exceptions, for example where there is editorial responsibility and human review of the content, or for artistic and satirical works, where disclosure must be given in a way that doesn't unduly hinder the work. There's also an exception for use that's legally authorised to detect, prevent, or investigate crime.

When and how the information has to be given

Whichever obligation applies, the information has to reach the person in a clear and distinguishable manner at the latest at the time of the first interaction or exposure — not afterwards. It also has to meet the applicable accessibility requirements.

Article 50 doesn't stand alone

It leaves the requirements for high-risk AI systems untouched, and it applies without prejudice to other transparency duties in EU or national law. Meeting Article 50 does not on its own mean you have met everything that applies to you.

What should your business consider in practice?

• Do you use chatbots or other AI systems that interact directly with customers?
• Do you generate content with AI that others might otherwise assume is human-made?
• Do you use AI for emotion recognition or biometric categorisation, in customer service or security, for example?
• Do you have processes in place to ensure AI-generated or manipulated content gets labelled or disclosed correctly?

This is exactly the kind of practical overview Complaia is also working to make easier to keep track of in the upcoming AI compliance platform.

Important to know

This page provides general information about Article 50 and is not legal advice. How the rules apply to a specific AI system depends on the precise circumstances, and we recommend seeking professional legal advice where it genuinely matters. Prepared and reviewed by Complaia based on Article 50 of the AI Act and the European Commission's guidelines of 20 July 2026.

The Complaia Guide: a deeper breakdown is on its way

Now that the rules apply, we're working on a focused, practical guide that goes even deeper on the transparency obligations specifically, written for businesses, not lawyers. It covers the transparency obligations only, not the full AI Act. It isn't ready yet, and it's general information rather than legal advice.

Want to be notified when the Complaia Guide is ready?