On 20 July 2026 the European Commission published its guidelines on the transparency obligations for providers and deployers of certain artificial intelligence systems: the obligations laid down in Article 50 of the AI Act, which apply from 2 August 2026. On the same day, the page of the Code of Practice on Transparency of AI-Generated Content gained a set of European icons for labelling, while the questions and answers on signing officially confirmed 27 July as the deadline for joining the first list of signatories. Ten days before the deadline, the toolbox is complete. The legal framework, now, is complete too: the Digital Omnibus on AI, the act that rewrites some of the rules the Commission already presents as settled, was published in the Official Journal of the European Union on 8 July 2026 as Regulation (EU) 2026/1744, in force from 27 July 2026.
What the guidelines clarify
The guidelines define the scope of the transparency obligations and are addressed not only to providers and deployers but also to the competent authorities called upon to supervise them: the stated aim is a consistent, effective, proportionate and uniform application across the Union. On the providers’ side, systems designed to interact directly with people must make it recognisable that one is dealing with a machine, and AI-generated or manipulated content must carry machine-readable marks enabling its detection. On the deployers’ side, information is owed when people are exposed to deepfakes, to AI-generated text published on matters of public interest without human review or editorial control, or to emotion recognition and biometric categorisation systems.
The practical value of the document lies in its definitions, exemptions and examples. The guidelines explain what constitutes a directly interactive system, the chatbot being the textbook example, what falls within synthetic content, including text generated wholly or partly by AI, and which interventions remain outside the scope, such as standard editing: correcting spelling and grammar does not turn a human text into content that must be marked.
The icons and the code: the practical route to demonstrating compliance
Alongside the guidelines, the Commission has made available a set of European icons that deployers may use to label AI-generated content: a voluntary tool designed to make labelling immediately recognisable. The code of practice remains structured in two sections, the first for providers, on marking and detection of content, the second for deployers, on labelling deepfakes and generated text; signatories will work together in dedicated taskforces to share implementation practices. On the technical side, machine-readable marking continues to rest on standards born outside Europe, an issue already examined on these pages with regard to the marking of content.
The difference between signing and not signing is one of evidential convenience, not of substance. Those who adhere to the code can rely on its measures to demonstrate compliance; those who choose other routes will have to demonstrate the adequacy of their measures case by case before the various market surveillance authorities, with the risk of diverging readings from one Member State to another, in a Union that is seeking a firmer legal basis precisely for dialogue between regulators. What the Commission wrote in its opinion of 8 July still stands: adherence to the code does not constitute conclusive evidence of compliance with the obligations of Article 50.
The calendar: 27 July, then 2 August
The extension announced by the AI Office by email, which we reported when reconstructing the signing calendar, is now official: to appear in the first list of signatories, to be published before 2 August, the form must be submitted by 18:00 CEST on 27 July 2026. Signing remains possible afterwards, on an ongoing basis. From 2 August, beyond the transparency obligations, the majority of the AI Act’s rules become applicable, including the enforcement powers of the Commission and of the national market surveillance authorities: the season in which content transparency was a matter of goodwill ends there.
The Digital Omnibus on AI is in the Official Journal: the 2 December deferral is law
In its press release the Commission writes that systems placed on the market before August will have to comply with the marking and detection obligations from 2 December 2026. That date is not in the AI Act: it comes from the Digital Omnibus on AI, approved by the European Parliament on 16 June and by the Council on 29 June 2026, and published in the Official Journal of the European Union on 8 July 2026 as Regulation (EU) 2026/1744, in force from 27 July 2026. By amending the AI Act, it introduces a four-month transitional period from 2 August 2026 for the marking obligation of Article 50(2), for providers that had already placed their systems on the market: hence the 2 December 2026 deadline. That deferral, until a few days ago a political landing point, is now an applicable rule.
For providers and deployers, the coming week is decisive in operational terms: those who wish to appear in the first list sign by 27 July; everyone else prepares by their own means the evidence of compliance, knowing that from 2 August the transparency obligations apply in the version currently in force. For pre-existing systems, then, the countdown now runs to 2 December 2026, when the marking and detection obligations become enforceable against them too.
