On 2 August 2026, the transparency and labelling obligations under Article 50 of the EU AI Act will enter into force. The obligations apply to providers and deployers of AI systems throughout Europe: chatbot providers, platform operators and medium-sized enterprises that publicly deploy AI-generated content alike. On 20 July 2026, the European Commission published the final Guidelines on Transparency Obligations for Providers and Deployers of AI Systems. They clarify what applies from now on, and provide the practical guidance that businesses have been waiting for.
Background: Article 50 AI Act and the Commission Guidelines
Article 50 of the AI Act (Regulation (EU) 2024/1689) requires providers and deployers of certain AI systems to inform users transparently when they are interacting with an AI system, or when content has been generated or altered by AI. The objective is to ensure that users are in a position to assess the nature of what they are communicating with and of the content being presented to them.
Following a public consultation period, the Commission published its final Guidelines in time for the date of application. Their stated aim is a coherent, effective and proportionate implementation of the requirements across all actors. The Guidelines are available at: https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems
Who Is Subject to These Obligations?
Article 50 AI Act addresses two categories of actors: providers (manufacturers or developers of AI systems) and deployers (organisations that deploy AI systems in their own context). The obligations differ depending on the role. Businesses may simultaneously act in both capacities and will in that case be subject to cumulative requirements.
A critical point: neither the Commission nor any other authority will proactively notify affected businesses. Assessing whether one’s own operations fall within scope is each business’s own responsibility. Failure to act entails regulatory risk. Pursuant to Article 99 AI Act, fines may reach EUR 15 million or 3 % of total worldwide annual turnover, whichever is higher.
The Four Regulatory Areas under Article 50 AI Act
a) Disclosure of AI-Driven Interactions (Article 50(1) AI Act – Provider Obligation)
Interactive AI systems such as chatbots, voice assistants, AI avatars and social media bots must be designed so that users clearly recognise they are not communicating with a human being. The relevant disclosure must be made no later than at the first point of contact and must be transparent, comprehensible and readily apparent. Particularly vulnerable groups, such as minors, must be given special consideration.
A disclosure obligation does not apply where the AI nature of the system is obvious to a reasonably well-informed user, for example in the case of AI coding assistants used by professionals or non-player characters in video games. It is not sufficient to place the disclosure solely in general terms and conditions, in the URL or in machine-readable form only.
b) Labelling of Synthetic Content (Article 50(2) AI Act – Provider Obligation)
Providers of AI systems that generate or manipulate images, videos, audio files or text must mark synthetic content in machine-readable form and at the same time ensure its recognisability for users. Since no single technique currently meets all requirements, the Guidelines recommend a combination of several approaches, in particular metadata, watermarks and visible indicators.
Exempt are purely assistive functions that do not materially alter content, such as automatic spell-checking, simple image rotation, noise reduction or technical format conversion. Subject to labelling obligations, by contrast, are AI-generated translations or summaries that add new content, background removal, the insertion of new image elements, or the creation of composite images.
Important note: For providers of GPAI systems (general-purpose AI, including large language models), the Omnibus Regulation provides for a transitional period until 2 December 2026 in respect of the technical labelling obligation under Article 50(2) AI Act.
c) Information Obligations for Biometric Analysis (Article 50(3) AI Act – Deployer Obligation)
Deployers of AI systems for emotion recognition or biometric categorisation must inform the data subjects concerned in advance and transparently of the use of such systems. This applies not only to dedicated AI providers but also to employers, platform operators and service providers using corresponding analytical tools. Integration into existing data protection consent procedures is permissible.
d) Deepfakes and Texts of Public Relevance (Article 50(4) AI Act – Deployer Obligation)
Deployers must disclose when content has been artificially generated or manipulated. A deepfake is defined as content that closely resembles a realistically possible person, place or event to a degree that a viewer could perceive it as authentic. According to the Guidelines, an intention to deceive is not required. Clearly excluded are depictions that are contrary to the laws of nature, such as flying humans or talking elephants.
Also subject to the labelling obligation are AI-generated texts intended to inform the public, such as AI-produced article summaries on websites, generated public warnings, or AI-manipulated corporate reports. An exception applies where human review and the assumption of editorial responsibility are present cumulatively. Creative, satirical or fictional works and purely commercial advertising without regulated claims are in principle exempt.
Accompanying Measures and Supporting Materials Published by the Commission
In parallel with the Guidelines, the Commission has published further materials that are of practical value for implementation:
- Code of Practice on AI-Generated Content: Code of conduct issued by the EU AI Office setting out technical and organisational details for labelling practice. Adherence is voluntary, but the Code is expected to become the de facto standard in supervisory proceedings (https://lnkd.in/epuEt_Vd).
- FAQs on Transparency Obligations: Practice-oriented answers from the Commission to frequently asked questions on the interpretation of Article 50 AI Act (https://lnkd.in/es9h8WJq).
- Quick Facts on Transparency Rules for AI Systems: Concise overview suitable for newcomers and for internal communication purposes (https://lnkd.in/eC-JdQih).
What Applies as of 2 August 2026, and What Does Not (Yet)?
The transparency obligations under Article 50 AI Act apply from 2 August 2026 to all systems within scope, including those already placed on the market before that date. According to the Guidelines, retrospective labelling of content already created or published is not required.
Exception: The technical labelling obligation for synthetic content under Article 50(2) AI Act is deferred for GPAI system providers until 2 December 2026. The remaining obligations under Article 50, in particular the disclosure obligation in the context of AI interaction and the deepfake labelling requirement, apply without exception from 2 August 2026.
Outside the scope of the obligations are: purely private uses without public dissemination, purely assistive AI functions that do not materially alter content, machine-to-machine communication, and internal industrial applications without user-facing interfaces.
Practical Tip:
2 August 2026 is no longer a distant deadline. Businesses that have not yet conducted an inventory of their AI systems should do so without delay. The review must cover not only the business’s own use of AI. Third-party service providers that create AI-generated content or operate AI systems on behalf of the business may also trigger obligations.
Recommended Immediate Actions:
- AI Systems Inventory: Identify all AI systems deployed or offered by your organisation and assess whether, and under which category of Article 50 AI Act, they fall within scope.
- Establish Role Clarity: Determine for each system whether your organisation is acting as a provider, a deployer or in both capacities. The obligations differ materially depending on the role.
- Implement Disclosure Mechanisms: Ensure that users receive a clear, comprehensible notice no later than at their first point of contact with an AI system. A disclosure included solely in terms and conditions or privacy policies is not sufficient.
- Labelling Technology for Synthetic Content: Assess whether your content-generating systems support machine-readable marking, and plan for the implementation of a combination of metadata and user-facing indicators.
- Review the Code of Practice: Consider whether signing the voluntary Code of Practice is strategically advisable for your organisation. Adherence may be advantageous in the event of supervisory proceedings.
- Internal Processes and Training: Raise awareness among marketing, communications and IT teams of the new requirements, and establish clear internal responsibilities for labelling compliance.
The AI Act is not a one-off compliance exercise. The requirements relating to transparency and labelling will continue to evolve in line with the technical development of AI systems and the further elaboration of the regulatory framework, in particular through the Code of Practice and future amendments under the Omnibus Regulation. We would be pleased to advise you on the development and implementation of a future-proof AI compliance strategy.
Contact:
Jens Borchardt