EU AI Act Article 50: Transparency Obligations Explained

Last updated: September 2, 2026

Article 50 of the EU AI Act establishes transparency obligations for providers and deployers of certain AI systems. It addresses direct interaction with AI, machine-readable marking of AI-generated content, emotion recognition and biometric categorisation, deepfakes, and certain AI-generated public-interest text.

Article 50 applies from August 2, 2026. It does not require every use of AI to be labelled in the same way. The applicable obligation depends on the organisation’s role, the type of AI system, the output, and how the system or content is presented to people.

Important legal notice

This page provides a general technical overview. It is not legal advice, does not determine whether Article 50 applies to a particular organisation or activity, and should not be treated as a compliance assessment.

The official regulation, European Commission guidelines, decisions of competent authorities, and applicable national or EU law take precedence over this overview. Organisations should obtain appropriate legal advice for their circumstances.

Providers and deployers have different responsibilities

The AI Act distinguishes between providers and deployers of AI systems.

A provider generally develops an AI system, has one developed, or places it on the market or puts it into service under its own name or trademark.

A deployer generally uses an AI system under its authority in a professional or organisational context. Personal, non-professional use is generally treated differently.

An organisation may be a provider for one system and a deployer for another. Its role can also depend on how a system is developed, branded, modified, integrated, or used.

1. Direct interaction with an AI system

Article 50(1) addresses AI systems intended to interact directly with natural persons, including certain chatbots, AI agents, and conversational interfaces.

Providers must design and develop these systems so that people are informed that they are interacting with an AI system, unless this would be obvious to a reasonably well-informed, observant, and circumspect person in the relevant circumstances.

The information must be clear and distinguishable and must be provided no later than the first interaction. Applicable accessibility requirements must also be considered.

A general statement on a separate transparency page may provide useful supporting context, but it may not replace a notice that is required at the point of interaction.

2. Machine-readable marking of AI-generated content

Article 50(2) applies to providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video, or text content.

Providers must ensure that relevant outputs are marked in a machine-readable format and can be detected as artificially generated or manipulated. The technical solutions must be effective, interoperable, robust, and reliable as far as technically feasible, taking account of factors including the type of content, implementation costs, technical limitations, and the generally acknowledged state of the art.

The obligation includes specific exceptions, including certain systems that perform an assistive function for standard editing or do not substantially alter the provided input or its meaning.

A visible website label, WordPress badge, HTML meta tag, disclosure notice, or transparency page is not necessarily equivalent to the provider-side machine-readable marking required by Article 50(2).

AIVOW Lite does not create or verify provider-side watermarks, C2PA credentials, durable provenance information, or other provider-level marking and detection mechanisms.

3. Emotion recognition and biometric categorisation

Article 50(3) requires deployers of emotion recognition or biometric categorisation systems to inform natural persons exposed to the operation of those systems.

The processing of personal data must also comply with applicable data-protection law. Other AI Act provisions may restrict or prohibit particular biometric or emotion-recognition uses independently of this transparency obligation.

AIVOW Lite does not detect whether a website uses emotion recognition or biometric categorisation and does not determine whether a particular system or use is permitted by law.

4. Deepfakes and public-interest text

Article 50(4) includes disclosure obligations for certain AI-generated or manipulated content.

Deepfakes

Deployers of AI systems that generate or manipulate image, audio, or video content constituting a deepfake must disclose that the content has been artificially generated or manipulated.

For evidently artistic, creative, satirical, fictional, or analogous works or programmes, the disclosure may be provided in an appropriate manner that does not unnecessarily interfere with the display or enjoyment of the work.

Machine-readable marking embedded by a provider does not necessarily replace a clear disclosure to the people exposed to the content.

Public-interest text

Deployers must disclose when AI-generated or manipulated text is published for the purpose of informing the public about matters of public interest.

The obligation does not apply where the content has undergone a process of human review or editorial control and a natural or legal person holds editorial responsibility for its publication.

According to the European Commission’s guidance, superficial checks such as spelling or grammar correction do not by themselves constitute the required substantive human review or editorial control.

Whether text concerns a matter of public interest and whether the review exception applies depend on the specific publication, subject, process, and allocation of editorial responsibility.

Timing, clarity, and accessibility

Under Article 50(5), required information must be presented in a clear and distinguishable manner no later than the first interaction or exposure.

This means organisations may need to consider:

Where the notice appears.
When the person first sees or hears it.
Whether the wording is understandable.
Whether it is visually or audibly distinguishable.
Whether applicable accessibility requirements are met.
Whether the notice remains appropriate across different devices and presentation formats.

A disclosure that is difficult to find or shown only after the relevant interaction or exposure may not fulfil the intended transparency function.

Application date and limited transition

Article 50 applies from August 2, 2026.

European Commission materials describe a proposed limited transition until December 2, 2026 for the Article 50(2) marking and detection obligation for certain generative AI systems placed on the market or put into service before August 2, 2026. The Commission notes that this transition depends on adoption of the relevant AI Omnibus amendment.

The Commission also states that content generated before August 2, 2026 does not have to be labelled retroactively under these requirements, although voluntary labelling is encouraged where possible.

Organisations should consult the current official sources because transitional rules, guidance, and implementation details may be updated.

How AIVOW Lite can help

AIVOW Lite provides a manual technical workflow for documenting AI systems and managing selected WordPress transparency outputs.

It can help administrators:

Maintain structured records of AI systems used by a website.
Record providers, purposes, data categories, external processing, retention, human oversight, public disclosures, and provider-marking information.
Manage systems as Active, Draft, or Inactive.
Display configurable AI Content Badges.
Show a Global AI Banner.
Display information about an active system through a Disclosure Shortcode.
Request acknowledgement through an Interaction Gate.
Generate a draft Public Transparency Page for review.
Add optional Lightweight AI Metadata to labelled content.
Review an informational Article 50 Readiness Matrix.
Maintain a separate AI Literacy Checklist.

These features can support internal documentation and public communication, but they do not establish or prove legal compliance.

What the plugin does not do

AIVOW Lite does not:

Automatically detect AI-generated or AI-manipulated content.
Determine whether an organisation is a provider, deployer, importer, distributor, or another regulated actor.
Decide whether a particular disclosure is legally required.
Classify an AI system as prohibited, high-risk, limited-risk, or otherwise regulated.
Verify provider compliance with machine-readable marking requirements.
Add durable provenance, C2PA credentials, or provider-side watermarking.
Verify that human review or editorial control satisfies a legal exception.
Detect emotion recognition or biometric categorisation systems.
Assess accessibility compliance.
Monitor regulatory changes automatically.
Provide legal advice or compliance certification.
Guarantee compliance with Article 50 or any other law.

The Interaction Gate records no legal conclusion and should not be treated as authentication, access control, valid consent, or proof of compliance.

Practical review questions

When reviewing a website or AI workflow, organisations may consider:

Which AI systems are being developed, provided, integrated, or used?
Is the organisation acting as a provider, deployer, or both for each system?
Does an AI system interact directly with natural persons?
Does it generate or manipulate synthetic audio, images, video, or text?
Does the provider supply appropriate machine-readable marking or detection mechanisms?
Is emotion recognition or biometric categorisation involved?
Could any published content constitute a deepfake?
Is AI-generated or manipulated text used to inform the public on matters of public interest?
Is any claimed human review substantive, documented, and subject to editorial responsibility?
Is required information presented clearly at the first interaction or exposure?
Have accessibility and other applicable legal requirements been reviewed?
Are the records and disclosures updated when the system or its use changes?

These questions are starting points for internal review and do not replace a legal assessment.

Article 50 is not the entire AI Act

Article 50 addresses specific transparency obligations. Other AI Act provisions may apply to prohibited practices, AI literacy, high-risk systems, general-purpose AI models, human oversight, documentation, record-keeping, risk management, or other matters.

Data-protection, consumer-protection, accessibility, intellectual-property, employment, sector-specific, and national laws may also apply independently.

Using a disclosure label or transparency plugin does not resolve those separate requirements.