Code of Practice on Transparency of AI-Generated Content (marcatura ed etichettatura dei contenuti generati o manipolati dall'IA)

Labelling deep fakes and AI-generated and manipulated published text applicable to deployers of AI systems (Article 50(4) and (5) AI Act)

-

Anja Bechmann Giovanni De Gregorio Madalina Botan Working Group 2 Chair Working Group 2 Vice -Chair Working Group 2 Vice -Chair

deployers of AI systems (Article 50(4) and (5) AI Act) Objectives The overarching objective of this Code of Practice (“Code”) is to improve the functioning of the internal market, to promote the uptake of human-centric and trustworthy artificial intelligence (“AI”), while ensuring a high level of protection of health, safety, and fundamental rights enshrined in the Charter, including democracy, the rule of law, and environmental protection, against harmful effects of AI in the Union, and to support innovation pursuant to Article 1(1) AI Act.

To achieve this overarching objective, the specific objectives of this Section of the Code are:

a) to serve as a guiding document for demonstrating compliance with the obligations of deployers of generative AI systems provided for in Article 50(4) and (5) AI Act, while recognising that adherence to the Code does not constitute conclusive evidence of compliance with these obligations under the AI Act;

b) to ensure that deployers of AI systems that generate or manipulate image, audio or video content constituting a deep fake, or text published with the purpose of informing the public on matters of public interest comply with their obligations under Article 50(4) and (5) AI Act, and

c) to enable the competent market surveillance authorities to assess compliance by deployers who rely on the Code to demonstrate compliance with those obligations under Article 50(4) and (5) AI Act in a consistent, predictable, and uniform manner across the Union.

Recitals Whereas:

a) Clear and distinguishable disclosure: Signatories recognise that advances in generative AI enable deployers of AI systems to generate or manipulate content that looks increasingly realistic, making it more difficult for people to distinguish it from authentic content. This highlights the need for transparency to protect public trust, safeguard democratic discourse, and inform individual decision-making. With the assistance of AI systems, deployers can generate or manipulate deep fakes that resemble existing persons, objects, places, entities or events, falsely appearing authentic or truthful to natural persons, as well as text on matters of public interest that is published without human review and editorial control. Generation and dissemination of such content pose specific risks of deception and manipulation, undermining trust in the information ecosystem and threatening its integrity. Signatories acknowledge that clear and distinguishable disclosure of the artificial origin or modification of such content required under Article 50(4) and (5) AI Act is a necessary safeguard to make individuals aware and inform their decision-making. This safeguard is not aimed at providing individuals with information about the trustworthiness of the content,

but rather at informing them about its artificial origin or the role of the AI system in manipulating the content.

b) Consistent and effective implementation: To make such disclosure clear and distinguishable for all natural persons, Signatories recognise the need for a certain level of uniformity in the disclosure methods as laid out in the Commitments of this Section of the Code. Such methods must be effective and aligned with the technical state of the art (including for user experience) to the extent feasible and appropriate for the context of dissemination. In particular, an optional EU icon provides an easy, practical and uniform pathway under the Code for compliance with Article 50(4) and (5) AI Act. At the same time, Signatories recognise the broad transversal scope of Article 50(4) AI Act as covering a multitude of use cases and contexts across sectors, media, platforms and modalities. Signatories acknowledge the need for a minimum level of uniformity and consistency regarding the implementation of the design and placement specifications for the icon or equivalent labels. At the same time, Signatories may further detail how to implement the measures envisaged in this Section of the Code in sector-specific good practices compliant with Article 50 AI Act and this Section of the Code.

c) Accessibility: Signatories recognise the importance of ensuring accessible disclosure of AI- generated or manipulated deep fakes and text published on matters of public interest within the scope of Article 50(4) AI Act, by taking into account the needs of different groups of vulnerable people. Disclosure should be designed to ensure it is easily perceivable and understandable by all natural persons, including persons with disabilities, children and the elderly.

d) Awareness and accountability: Signatories recognise that internal processes, literacy and awareness measures, and review mechanisms are essential for ensuring effective disclosure of AI-generated or manipulated deep fakes and text published on matters of public interest within the scope of Article 50(4) AI Act. To achieve this, Signatories commit to label such AI- generated and manipulated content and to disclose it at the latest at the first exposure based on the design and placement specifications in this Code. Considering Signatories’ available organisational resources and technical capacities, this also includes proportionate internal processes and measures which aim to ensure that individuals are informed about the artificial nature of the content to which they are exposed, including by making publicly available information about the use by the Signatory of the EU icon or equivalent labels and their meaning.

e) Promoting an ecosystem of transparency and innovation: Signatories acknowledge that compliance with Article 50(4) AI Act is not a check-box exercise but aims to create an ecosystem of transparency that ties into organisational practices, technological developments, and individuals’ awareness. It has the shared goal of empowering individuals to make informed decisions by enabling them to understand the origin and nature of synthetic content. In such an ecosystem, Signatories recognise that the Code allows for continuous developments in a fast-changing technological landscape and in light of the transversal scope of Article 50(4) AI Act. Key factors in such a structure of the Code are to build an ecosystem that preserves uniformity and recognisability, while allowing for new evidence to be gathered and adapted into good practices that support the Code and compliance with Article 50(4) and (5) AI Act.

f) Additional safeguards under other Union and national law: Signatories acknowledge that the transparency obligations under the AI Act apply in parallel to, and do not affect, other legal obligations or requirements that may apply to the creation, distribution or use of AI- generated or manipulated deep fakes and text published on matters of public interest under applicable Union legislation (e.g., on data protection, consumer protection, digital services, intellectual property, media law, political advertising, criminal law and other relevant regulatory frameworks).

Commitments This Section of the Code applies only to Signatories in so far as they are deployers of AI systems that generate or manipulate image, audio or video content constituting a deep fake1 or text published with the purpose of informing the public on matters of public interest falling within the scope of Article 50(4) AI Act. Each reference below to “content”, “deep fake” or “published text” implies content that is i) AI-generated or manipulated and that qualifies as a deep fake under Article 3(60) AI Act (referred to as “deep fake”), or ii) AI-generated or manipulated text published with the purpose of informing the public on matters of public interest, without human review or editorial control and where no natural or legal person holds editorial responsibility for the publication of the content (referred to as “published text” or “text published on matters of public interest”).

Considering the diversity of deployers and contexts in which deep fakes and published text may be created and disseminated, Signatories and the associations of which they are part may jointly establish common good practices for their sector or industry, which could provide further detail regarding implementation of the Commitments and Measures specified in this Section of the Code. Any such good practices should be fully aligned with the measures set out below, without preventing Signatories from tailoring the disclosure practices to their specific sectors to account for sector specificities and existing rules, nor from providing additional transparency, where relevant. To ensure transparency, accountability and to enable monitoring by competent market surveillance authorities, sectoral good practices established in accordance with this paragraph will be made publicly available online.

In this Section, the terms “will”, “encouraged” and “may” should be interpreted in the following way:

1. ‘will’ is used for mandatory measures under the Code that need to be met to be compliant with Article 50(4) or (5) AI Act and that will be monitored by competent market surveillance authorities;

2. ‘encouraged’ is used for optional measures under the Code that are not legally required and are purely voluntary but are nevertheless recommended;

3. ‘may’ is used for optional measures under the Code that are not legally required and are purely voluntary or provide deployers with flexibility on how to implement the respective measure or commitment.

A deep fake is defined in Article 3(60) AI Act as 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.