Disclosing AI-Generated Content Under the EU AI Act
In this blog post, we examine the transparency and disclosure requirements for AI-generated and AI-manipulated content under Article 50 of the EU AI Act. We discuss which types of content require disclosure, how those disclosures should be presented, and how the European Commission's Guidelines and the Code of Practice further specify these requirements. Finally, we introduce vera[tag] and Kladero's broader approach to supporting AI-content disclosure and transparency workflows.
Context
Generative artificial intelligence (“GenAI”) tools, such as ChatGPT, Claude, Gemini, Canva, Adobe Firefly, and CapCut, are increasingly used to create online content across digital media, including social media, blogs, and commercial websites. The use of GenAI tools has also become increasingly common for creating content for productive and commercial purposes, such as the generation of promotional and marketing material, advertisements, websites for products and services, and product descriptions for online shops [1], [2].
However, as the outputs of GenAI (“AI-generated content” or, depending on how the tool was used, “AI-manipulated content”) become increasingly difficult to distinguish from human-created content [3], concerns have emerged regarding the risks of misinformation, manipulation, and deception for online audiences. These concerns have prompted a growing number of regulatory requirements and platform-specific rules addressing the transparency of AI-generated content. Notably, the European Union's Artificial Intelligence Act (“AI Act”) establishes a set of transparency obligations concerning the disclosure of AI-generated and AI-manipulated content.
What Should Be Disclosed
Article 50 of the AI Act, in force since 2 August 2026, sets out transparency obligations concerning certain AI-generated and AI-manipulated content [4]. Under Article 50, the relevant disclosure obligations apply to individuals or organizations that use AI systems in a professional context, referred to in the AI Act as deployers of AI systems. These obligations are particularly relevant to marketers, publishers, agencies, and other digital content creators that use GenAI tools to create or modify online content. According to the European Commission, the disclosure obligations apply to AI-generated or AI-modified content made available from 2 August 2026 onward [5], [6].
In Article 50, the EU AI Act addresses two main types of AI-generated or AI-manipulated content: deep fakes and text informing on matters of public interest, both discussed below.
Deep Fakes
Deployers must disclose when AI has been employed in the generation or manipulation of image, audio, or video content, if that content constitutes a deep fake. Article 3(60) of the EU AI Act defines deep fakes as follows:
“‘deep fake’ means AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful;”
Further analysis of what would constitute deep fakes, together with examples, are discussed in guidelines published by the European Commission on the implementation of the transparency obligations [7]. In particular, the guidelines further examine the notion of deep fakes from the perspective of four criteria:
-
Resemblance. There should be a high degree of similarity between the AI-generated or AI-manipulated content and the subject it resembles.
-
Existing. The subject represented by the AI-generated or manipulated content should appear realistic rather than obviously unrealistic or simulated (e.g., by defying the laws of nature or physics).
-
Persons, objects, places, entities, or events. The subject of the AI-generated or AI-manipulated content should be a person (including a real person or realistic digital avatar), object, place, entity (including a realistic animal or other biological lifeform), or event (including a realistic scene involving persons, objects, places, or entities).
-
False appearance to a person to be authentic or truthful. The AI-generated or manipulated content should be capable of giving a person the false impression that the content is authentic or truthful, for example, through its photorealism.
Whether these criteria are met for a specific AI-generated or AI-manipulated content should be assessed objectively by the deployer, taking into account the content, its context of use, and the reasonably foreseeable audience [7].
Examples of deep fakes may include cloned voices, synthetic depictions of politicians or public figures, and AI-generated product images that may mislead audiences about the actual product. By contrast, clearly fantastical scenes may fail the existing criterion, while minor technical edits or AI-generated backgrounds may fall outside the definition where they do not affect the audience's perception of the content as authentic or truthful [7].
Text Informing on Matters of Public Interest
In addition to deep fakes, deployers must disclose any AI-generated or AI-manipulated text published with the purpose of informing the public on matters of public interest (cf. Article 50(4), second subparagraph of the EU AI Act).
By published it can be understood that the text is accessible to the wide public, regardless of whether it requires access payment and not available only to a small, closed, or private group of individuals [7].
Published should be understood as meaning that the text is made accessible to the wide public, irrespective of whether access requires payment, rather than being available only to a small, closed, or private group of individuals [7]. The purpose of informing the public implies that the text is intended to communicate knowledge, opinions, or facts [7].
The scope of matters of public interest is less fixed and may evolve over time [7]. The European Commission's guidelines indicate that the concept should encompass matters relevant to society at large, including “politics and democratic processes, public administration and services, the administration of justice and law enforcement, the protection of fundamental rights, public security, public health, environmental protection, consumer safety, and any economic, financial, political, scientific, or cultural development that may be relevant subject of public debate” (cf. [7]).
Examples of AI-generated or AI-manipulated text falling within the category of text informing on matters of public interest include AI-generated summaries of newspaper articles discussing recent policy changes, AI-manipulated articles or portions of articles comparing the effects of diets on diseases, and AI-generated messages concerning extreme-weather warnings (cf. [7]). By contrast, AI-generated or AI-manipulated texts such as fantasy novels or product advertisements that do not contain health or safety claims would generally not be regarded as informing the public on matters of public interest [7] and would therefore not be subject to this disclosure obligation.
Note that, in cases where (1) the AI-generated or AI-manipulated text has undergone human review or editorial control and (2) a natural or legal person holds editorial responsibility for the publication, then the respective AI-generated or AI-manipulated text is not subject to disclosure obligations (cf. Article 50(4), second subparagraph). According to the European Commission's guidelines, the human review should include, at a minimum, fact-checking aimed at verifying the accuracy of the information [7].
How Should AI-Generated or AI-Manipulated Content be Disclosed
Three main sources set out the requirements and guidance for disclosing AI-generated or AI-manipulated content: (1) the EU AI Act, in particular Article 50; (2) the European Commission's Guidelines [7]; and (3) the Code of Practice on Transparency of AI-Generated Content [8]. The discussion below focuses on selected requirements and recommendations from these sources.
Article 50 of the EU AI Act
Article 50 of the EU AI Act sets out the general requirements for disclosing AI-generated and AI-manipulated content. Specifically (cf. Article 50(5)):
- The disclosure information must be provided in a clear and distinguishable manner.
- The disclosure information must be provided to the persons exposed to the AI-generated or AI-manipulated content no later than the time of their first exposure.
- The disclosure information must be provided in a manner that conforms with the applicable accessibility requirements.
In addition, where AI-generated or AI-manipulated content forms part of an evidently artistic, creative, satirical, fictional, or analogous work, Article 50(4) provides that the disclosure should be made in a manner that does not hamper the display or enjoyment of the work.
European Commission's Guidelines
The European Commission's Guidelines [7] provide further interpretation of Article 50 of the EU AI Act and set out more concrete guidance for disclosing AI-generated or AI-manipulated content, accompanied by practical examples. Specifically, the European Commission's Guidelines introduce the following refined requirements.
-
Clear and distinguishable. For disclosure information to be clear and distinguishable, the European Commission's Guidelines indicate that it should be noticeable, easy to understand, accessible, and readily identifiable as distinct from other information or from the surrounding presentation of the content. The disclosure should also take into account the reasonably foreseeable audience of the content, including children and persons with disabilities where relevant, and should not be easily overlooked under normal conditions of exposure.
-
First exposure. According to the European Commission's Guidelines, appropriate disclosure at the time of first exposure may include, for example, disclosure at the start of a deep fake video, at the beginning of AI-generated or AI-manipulated text, or when a user first encounters a deep fake while scrolling through social media. The Guidelines also recognize that persons exposed to AI-generated or AI-manipulated content may not always encounter the content from its beginning and may therefore miss an initial disclosure. For this reason, the Guidelines recommend supplementing the initial disclosure with additional disclosures at later points of exposure.
-
Accessible. For disclosure information to be accessible, the European Commission's Guidelines state that, where applicable, it should comply with Directive 2016/2102 on the accessibility of websites and mobile applications [9] and Directive 2019/882, the European Accessibility Act [10], concerning the accessibility of products and services.
Code of Practice on Transparency of AI-Generated Content
To support compliance with the AI Act's transparency obligations, the European Commission facilitated the development of the Code of Practice on Transparency of AI-Generated Content [8] (hereinafter, the Code of Practice). Drafted by independent experts and published on 10 June 2026, the Code of Practice provides concrete practical guidance on how AI-generated or AI-manipulated content should be disclosed so as to adequately inform persons exposed to such content.
Beyond providing practical guidance, the Code of Practice also serves as a benchmark against which a deployer's compliance with the EU AI Act's transparency obligations may be assessed. In particular, the European Commission's Guidelines state that deployers should carry out a gap analysis comparing the measures they have implemented with those set out in the Code of Practice. Where a deployer is a signatory to the Code of Practice, compliance may instead be assessed by examining the deployer's adherence to the Code of Practice (see paragraphs (146)–(148) of [7]).
The Code of Practice supports the compliance with the EU AI Act transparency obligations by providing practical guidelines across four main dimensions discussed next.
Design
The Code of Practice provides concrete design specifications for the disclosure of AI-generated and AI-manipulated content. To account for different content formats and presentation contexts, the Code of Practice distinguishes between cases in which a visual disclosure is possible (e.g., content displayed on a website) and cases in which visual disclosure is not possible (e.g., audio-only content).
-
Visual disclosure is possible. Where visual disclosure is possible, the Code of Practice provides for the use of an icon or equivalent label displaying the acronym “AI,” with both letters of equal height, unless national law requires the use of an equivalent acronym in the relevant national language. Depending on the context, the icon or equivalent label may be presented in different styles or sizes, provided that it remains readable and recognizable.
To provide further information about which elements of the content have been AI-generated or AI-manipulated, the icon or equivalent label may be accompanied by additional information displayed next to or within the icon, or presented through an interactive layer (e.g., a hover-triggered popup).
To facilitate consistent implementation of disclosures based on these design specification, the Code of Practice provides a concrete, free-to-use icon design, available for download here.
-
Visual disclosure is not possible. Where visual disclosure is not possible, the Code of Practice specifies that disclosure should be made through an audible disclaimer. The disclaimer should be brief, expressed in simple language, and delivered either in the same language as the content or in English. It should indicate the AI-generated or AI-manipulated origin of the audio content. To provide further information, the disclaimer may also include details about which elements of the content have been generated or manipulated by the AI system.
Placement
Placement concerns where and how the disclosure is presented in relation to the AI-generated or AI-manipulated content, so that it remains clear and distinguishable and is provided no later than the time of first exposure. According to the Code of Practice, disclosure placement should follow several general principles:
- Disclosures should be immediately recognizable without requiring persons to perform additional actions or sustain prolonged attention.
- Disclosures should remain perceptible for a sufficient duration to allow persons exposed to the content to notice and understand them.
- Disclosures should be directly embedded in the AI-generated or AI-manipulated content, unless equivalent measures are used (e.g., visual overlays displayed on the content). The placement of the disclosure should also take into account the distribution and dissemination chain through which the content may be made available.
- Disclosures should be sufficiently separated from other disclaimers, overlays, and visual or audio elements so that they remain distinguishable.
Measure 1.2 of the Code of Practice (pages 31–33 of [8]) sets out more detailed placement specifications based on these general principles. These specifications take into account, among other factors, the type of AI-generated or AI-manipulated content (e.g., text, video, or image), whether a screen is available, and the reasonably foreseeable audience or context of use (e.g., internal training materials for employees).
Accessibility
To ensure that disclosures of AI-generated or AI-manipulated content are accessible to a broad range of users, the Code of Practice emphasizes compliance with applicable accessibility legislation, in particular Directive (EU) 2019/882 [10] and Directive (EU) 2016/2102 [9], as well as with established accessibility standards such as the W3C Web Content Accessibility Guidelines (WCAG) 2.1 [11]. Specifically, the Code of Practice requires at least the following:
- Visual disclosures must be accompanied by audio descriptions or equivalent alternative cues.
- Audio-only content and associated audio disclosures must be accompanied by tactile or haptic cues to accommodate persons with hearing disabilities.
- Icons or equivalent labels must provide sufficient contrast to accommodate persons with color vision deficiencies.
- Disclosures must be detectable by assistive technologies and, where applicable, compatible with screen readers.
Internal Processes
To fulfill and demonstrate compliance with the EU AI Act's transparency obligations, the Code of Practice specifies a set of internal processes that deployers should establish, implement, and maintain. These include:
-
Documentation. Deployers must document how the disclosure obligations have been implemented, including through a general description of the measures adopted and representative examples of disclosures. In addition, deployers should publish, preferably publicly, descriptions of the icons, labels, or other disclosure mechanisms used, their meanings, and any relevant variations across different contexts.
-
Verification. Deployers must establish and maintain processes to mitigate the risk of AI-generated or AI-manipulated content being left unlabeled or labeled incorrectly. Such processes may include, for example, verifying that the applicable design and placement specifications have been implemented correctly.
-
Awareness and literacy. Deployers must take appropriate measures to ensure that personnel involved in implementing disclosure measures are sufficiently aware of the EU AI Act's transparency obligations. This may include providing training or guidance on identifying content that requires disclosure, addressing relevant accessibility considerations, and correcting missing or inaccurate disclosures. Personnel should also receive general AI-literacy training appropriate to their roles and responsibilities.
-
Feedback and review. Deployers must establish internal and external feedback processes. In particular, they should provide reporting channels through which persons and third parties can flag missing or incorrect disclosures. Sufficiently substantiated reports should be reviewed without undue delay, and any necessary corrective measures should be taken accordingly.
In addition, without prejudice to media freedom, editorial independence, and the protection of journalistic sources, deployers should maintain appropriate processes for human review and editorial control prior to the publication of AI-generated or AI-manipulated content. An identifiable natural or legal person should hold editorial responsibility for the publication. The review processes and the human resources allocated to them should be documented at an appropriate level of detail, and the contact details of the person responsible for ensuring editorial accountability should be made publicly available.
How vera[tag] Can Help
If you are a content creator or publisher dealing with large volumes of AI-generated content and increasingly complex disclosure requirements, vera[tag] by Kladero is designed for you. Built around the practical challenges of AI transparency, vera[tag] provides the tools needed both to present disclosures to your audience and to manage those disclosures over time.
vera[tag] was designed from the outset with the EU AI Act's transparency obligations in mind, together with the transparency guidelines issued by the European Commission and the concrete disclosure specifications set out in the Code of Practice.
With a vera[tag] subscription, you get a comprehensive set of tools for managing AI-content disclosures across their full lifecycle:
- Tools for seamlessly disclosing AI-generated and AI-manipulated content across a range of channels, including websites, blogs, and social media.
- Highly configurable visual disclosure components, with several ready-to-use styles and themes, including the icon set provided by the Code of Practice and vera[tag]'s own custom disclosure design.
- Flexible disclosure information that can be adapted to different use cases while retaining sensible defaults for common scenarios.
- Context-aware, page-style-adaptive, and interactive disclosure presentation, including overlays and other presentation mechanisms designed to integrate naturally with the surrounding content.
- Disclosure information presented in the user's language, helping make disclosures understandable across multilingual audiences.
- Accessible and machine-readable disclosure presentation, including support for assistive technologies such as screen readers and structured disclosure information that can be processed programmatically.
- Comprehensive documentation for both your team and end users, explaining how disclosures are configured, presented, and interpreted.
- Multi-account and team-management functionality supporting traceability and accountability across the disclosure workflow.
- … and much more.
Taken together, these features and capabilities make vera[tag] one of the best solutions for managing and delivering AI-content disclosures consistently and at scale. At Kladero, we continuously improve vera[tag] with new features, integrations, and support for evolving disclosure requirements across platforms and jurisdictions. Get started with vera[tag] today to simplify your disclosure workflow and spend more time creating and publishing content.
How Kladero Can Help
Beyond vera[tag], Kladero's mission is to help you address the AI transparency and disclosure requirements specific to your needs. Through custom solutions, consulting, and targeted training, Kladero's experts can help you interpret applicable requirements, design effective disclosure processes, and implement them across your existing content creation and publication workflows. Get in touch with Kladero today and take your content disclosure workflows to the next level.
Summary
In this blog post, we reviewed the transparency and disclosure requirements for AI-generated and AI-manipulated content under Article 50 of the EU AI Act. We discussed which types of content require disclosure and how the European Commission's Guidelines and the Code of Practice further specify the design, placement, accessibility, and internal processes associated with those disclosures. Finally, we described how vera[tag] can support the presentation and management of AI disclosures, and how Kladero can help address broader transparency and disclosure needs through custom solutions, consulting, and training.
Stay tuned for more posts on AI transparency, disclosure requirements, and practical approaches to implementing them.
Bibliography
Meet the author
