Nucleo di vetro luminoso da cui partono un'onda sonora, un'immagine e un flusso di dati, ciascuno attraversato da un anello di luce con un sigillo geometrico che indica la marcatura dei contenuti generati dall'IA

Immagine creata con IAAI-generated image

Transparency of AI-generated content: the guidelines on Article 50 of the AI Act and the 2 December 2026 deadline

On 12 November 2026 the European Commission published in the Official Journal of the European Union its guidelines on the transparency obligations under Article 50 of Regulation (EU) 2024/1689 on artificial intelligence (AI), known as the AI Act (Communication C/2026/6005). The document interprets obligations that have applied since 2 August 2026 and accompanies the 2 December 2026 deadline set by the Digital Omnibus on AI for marking content generated by systems already on the market.

The four obligations under Article 50

Article 50 of the AI Act lays down four transparency obligations, each addressed to a specific actor. Providers of systems intended to interact directly with natural persons, such as chatbots, must design them so that people are informed that they are interacting with an AI system, unless this is obvious to a reasonably well-informed, observant and circumspect person (paragraph 1). Providers of systems generating synthetic audio, image, video or text content, including general-purpose AI systems, must ensure that the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated, using technical solutions that are effective, interoperable, robust and reliable as far as technically feasible (paragraph 2).

The other two obligations concern deployers, that is, those using AI systems under their own authority. Deployers of emotion recognition or biometric categorisation systems must inform the persons exposed to them (paragraph 3). Deployers generating or manipulating image, audio or video content constituting a deep fake must disclose that the content has been artificially generated or manipulated; the same applies to text published with the purpose of informing the public on matters of public interest, unless it has undergone human review or editorial control and a natural or legal person holds editorial responsibility for it (paragraph 4).

Article 3, point 60, defines a deep fake 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. For evidently artistic, creative, satirical or fictional works, the obligation is limited to disclosing the existence of generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work. The information must be provided in a clear and distinguishable manner at the latest at the time of the first interaction or exposure and must conform to the applicable accessibility requirements (paragraph 5). Infringements may lead to administrative fines of up to EUR 15 million or, if higher, up to 3% of total worldwide annual turnover (Article 99(4)).

The Commission guidelines: structure and content

The guidelines are adopted on the basis of Article 96(1)(d) of the AI Act. The Commission had made them available on its website on 20 July 2026; the version published in the Official Journal runs to 43 pages. The document is not binding: the authoritative interpretation of the Regulation rests with the Court of Justice of the European Union.

A horizontal part sets out the purpose of the obligations (countering impersonation, deception, disinformation and fraud), the responsible actors, including providers and deployers established outside the Union, the scope of the exclusions (purely personal non-professional use, research and development) and the limits of the exemption for systems released under free and open-source licences, which under Article 2(12) does not apply to systems subject to Article 50 and the relationship with prohibited practices, high-risk systems, AI literacy and general-purpose AI models. This is followed by a section on each paragraph of Article 50 and a section on information requirements.

Among the clarifications, the exception for obvious interaction is to be interpreted narrowly, taking into account the target audience, including potentially vulnerable groups. Synthetic content requires both machine-readable marking and detectability, and meeting only one of the two is not sufficient; source code, machine-to-machine communications and internal processing are among the cases excluded. On information, the guidelines state that a notice contained only in general terms and conditions is not sufficient, and recommend accessible formats, in particular for persons with disabilities and vulnerable groups such as children and older people, as well as multimodal approaches (text, audio, visual cues). EU rules on consumer protection, commercial practices and data protection continue to apply.

The code of practice on marking and labelling

The guidelines complement the Code of Practice on marking and labelling of AI-generated content, published by the Commission on 10 June 2026. The code is voluntary and was drafted by six independent experts with input from more than 180 stakeholders, including providers and deployers, small and medium-sized enterprises (SMEs), academia, the public sector and civil society organisations. It is structured in two sections: the first, for providers, covers machine-readable marking of content; the second, for deployers, covers the labelling of deep fakes and of public-interest text without human review or editorial control.

The code is open for signature. According to the Commission, once the code has been assessed as adequate by the Commission and the European Artificial Intelligence Board (AI Board), signatories can use it to demonstrate compliance with the relevant obligations. The guidelines clarify the scope of the legal obligations and the aspects not covered by the code.

The Digital Omnibus on AI and the new timetable

Regulation (EU) 2026/1744 of 8 July 2026 (Digital Omnibus on AI), in force since 27 July 2026, amended the AI Act. For systems placed on the market before 2 August 2026, the transitional period for implementing the marking solutions required by Article 50(2) ends on 2 December 2026; recital 38 of the Regulation provides for a four-month transitional period for providers that had already placed their generative AI systems on the market before 2 August 2026.

The same Regulation adds to Article 5 a new prohibition on AI systems that generate or manipulate realistic images, videos, audio or similar material of an identifiable natural person’s intimate parts, or of an identifiable natural person engaged in sexually explicit activities, without that person’s freely given, specific, informed, unambiguous and explicit consent, as well as child sexual abuse material within the meaning of Directive 2011/93/EU; under Article 113, the prohibition applies from 2 December 2026. For high-risk systems, the application dates become 2 December 2027 for stand-alone systems under Annex III and 2 August 2028 for systems embedded in products under Annex I.

Outlook

2 December 2026 is the deadline for marking the outputs of generative systems already on the market. Article 50(7), as replaced by the Digital Omnibus, provides that the Commission assesses the adequacy of codes of practice for paragraphs 2 and 4, taking utmost account of the Board’s opinion, and may adopt an implementing act only if it deems a code inadequate; according to recital 41, codes have limited legal effect and do not grant a presumption of conformity. Still to be defined are enforcement in practice by national authorities (in Italy, Article 20 of Law No 132 of 23 September 2025 designated the Agency for Digital Italy, AgID, as notifying authority and the National Cybersecurity Agency, ACN, as market surveillance authority, without prejudice to the competences of the Bank of Italy, CONSOB and IVASS), the uptake of interoperable technical marking solutions, and any interpretation by the Court of Justice of the notions of deep fake and editorial control.

In dialogue with the 2030 Agenda

  • Goal 16. Peace, justice and strong institutions. Target 16.10 concerns public access to information and the protection of fundamental freedoms; marking and labelling obligations affect how recognisable the content circulating in the information space is.
  • Goal 5. Gender equality. Target 5.b concerns the use of enabling technology, in particular information and communications technology, to promote the empowerment of women; the new prohibition concerns intimate or sexually explicit AI-generated content produced without the consent of the person depicted.
  • Goal 9. Industry, innovation and infrastructure. Target 9.5 concerns enhancing scientific research and upgrading the technological capabilities of industrial sectors; the requirement for effective, interoperable, robust and reliable marking solutions presupposes shared technical developments among providers.

Sources


AI AnthropoCosmic In evidenzaAI AnthropoCosmicUn progetto internazionale per un’IA a servizio dell’Uomo, dell’Ambiente e del Cosmo, che mette al centro la dignità della persona nella progettazione dei sistemi. Leggi l’articoloAI Open Mind AI AnthropoCosmic FeaturedAI AnthropoCosmicAn international project for an AI at the service of humanity, the environment and the cosmos, placing human dignity at the centre of system design. Read the articleAI Open Mind Agentic AI In evidenzaAgentic AILimiti prima dell’azione, evidenze durante, responsabilità dopo. Il volume di Nicola Fabiano sulla governance dei sistemi agentici, con la prefazione di Antonino Caffo.Capitolo 16 a cura dell’Avv. Valentina Grazia SapuppoLeggi l’articolo Agentic AI FeaturedAgentic AILimits before the action, evidence during, responsibility afterwards. Nicola Fabiano’s book on the governance of agentic systems, with a preface by Antonino Caffo.Chapter 16 by Valentina Grazia SapuppoRead the article
Testimonianza di Valentina Grazia Sapuppo per XUKRA, Meeting Standards 2026 TestimonianzaXUKRA” · Meeting StandardsValentina Grazia Sapuppo ha portato la sua testimonianza al workshop online di XUKRA” nell’ambito di Meeting Standards, l’iniziativa di Small Business Standards che avvicina le PMI europee alla normazione.30 giugno 2026, online · Meeting Standards, 4ª edizioneXUKRA”Meeting StandardsIniziativa europea di Small Business Standards - SBS
Locandina del workshop DTRIP4H Exploring User Needs, 26 marzo 2026PanelDTRIP4H · Gemelli digitali per la saluteValentina Grazia Sapuppo tra i main panellist del workshop «Exploring User Needs for DDTE: Ethics, Implementation, and the Future of Responsible Simulation» del progetto europeo DTRIP4H, con un intervento su governance, conformità e responsabilità. A giugno ha contribuito come esperto legale al workshop sui fabbisogni formativi.26 marzo e 23 giugno 2026, online · Horizon EuropeIl workshopLa registrazioneDTRIP4H
Intervista Radio Radio IntervistaLegge e colossi del digitaleIl patteggiamento di Meta sui minori non è una condanna. Stati Uniti ed Europa seguono strade opposte, e sugli agenti di IA resta aperta la domanda su chi risponde.Un Giorno Speciale su Radio Radio, con Alessio De Paolis · audio dal minuto 2:26:00Ascolta l’intervistaGuarda il videoLeggi l’articolo
Digital Omnibus ContributoIl Digital Omnibus cambia l’AI ActNuove scadenze per i sistemi ad alto rischio e un chiarimento sull’obbligo di AI literacy: più tempo per adeguarsi, nessuno sconto sulla preparazione di persone e processi.Articolo scritto per il blog di SkillaLeggi su Skilla
Digeat Festival 2026 SpeakerDigeat Festival 2026Valentina Grazia Sapuppo tra i relatori del festival dedicato a protezione dei dati, archivi digitali e regole del futuro. Interviene sul tema «Le regole dell’IA: nuove leggi o principi del diritto?».Venerdì 6 novembre 2026, ore 16:30, Ex Convitto Palmieri, LecceL’interventoLa scheda relatriceIl festival
AI AnthropoCosmic 2026AI AnthropoCosmic 2026Valentina Grazia Sapuppo nel progetto dell’Università Pontificia Salesiana su Persona, Ambiente e Cosmo: moderazione della sessione mattutina del Convegno finale e intervento negli AI Laboratori del Domani su commercio elettronico e IA.14 novembre 2026, online · 28 novembre 2026, Università Pontificia Salesiana, RomaIl convegnoL’incontroIl contributoIl comitato scientificoLa relatriceIl progetto
Testimony by Valentina Grazia Sapuppo for XUKRA, Meeting Standards 2026 TestimonyXUKRA” · Meeting StandardsValentina Grazia Sapuppo’s testimony for the XUKRA” online workshop, part of Meeting Standards, the Small Business Standards initiative bringing European SMEs closer to standardisation.30 June 2026, online · Meeting Standards, 4th editionXUKRA”Meeting StandardsEuropean initiative by Small Business Standards - SBS
Poster of the DTRIP4H workshop Exploring User Needs, 26 March 2026PanelDTRIP4H · Digital twins for healthValentina Grazia Sapuppo among the main panellists of the workshop “Exploring User Needs for DDTE: Ethics, Implementation, and the Future of Responsible Simulation” of the EU project DTRIP4H, speaking on governance, compliance and accountability. In June she contributed as a legal expert to the workshop on training needs.26 March and 23 June 2026, online · Horizon EuropeThe workshopThe recordingDTRIP4H
Interview Radio Radio InterviewLaw and the digital giantsThe Meta settlement on minors is not a conviction. The United States and Europe take opposite paths, and on AI agents the question of who answers remains open.Un Giorno Speciale on Radio Radio, with Alessio De Paolis · audio from 2:26:00 · in ItalianListen to the interviewWatch the videoRead the article
Digital Omnibus ContributionThe Digital Omnibus reshapes the AI ActNew deadlines for high-risk systems and a clarification on the AI literacy duty: more time to comply, no discount on preparing people and processes.Article written for the Skilla blog, in ItalianRead on Skilla
Digeat Festival 2026 SpeakerDigeat Festival 2026Valentina Grazia Sapuppo among the speakers of the festival on data protection, digital archives and the rules of the future. She takes part in the panel «The rules of AI: new laws or principles of law?».Friday 6 November 2026, 16:30, Ex Convitto Palmieri, LecceThe panelSpeaker profileThe festival
AI AnthropoCosmic 2026AI AnthropoCosmic 2026Valentina Grazia Sapuppo in the project of the Università Pontificia Salesiana on Person, Environment and Cosmos: moderator of the morning session of the closing conference and speaker at the AI Laboratori del Domani on e-commerce and AI.14 November 2026, online · 28 November 2026, Università Pontificia Salesiana, Rome · sessions held in ItalianThe conferenceThe sessionThe contributionThe scientific committeeSpeakerThe project