On 8 July 2026 the European Commission concluded that the Code of Practice on Transparency of AI-generated content adequately covers the obligations laid down in Article 50(2), (4) and (5) of the AI Act and facilitates their effective implementation. The following day the AI Board adopted its own adequacy assessment. In the very same weeks, however, the final text of the Digital Omnibus on AI, approved by the European Parliament on 16 June and by the Council on 29 June 2026, intervenes precisely on Article 50: it pushes back the marking obligation for systems already on the market, and it removes the Commission’s power to approve that code by way of an implementing act, on a ground worth reading in full. Codes of practice, recital 41 states, have limited legal effects and, in particular, do not confer a presumption of conformity.
What applies from 2 August, and what slips to 2 December
Article 50 of the AI Act distinguishes two sets of addressees. Providers of generative AI systems must mark their outputs, whether audio, image, video or text, in a machine-readable format and make them detectable as artificially generated or manipulated, through technical solutions that are effective, interoperable, robust and reliable as far as this is technically feasible. Deployers must disclose that content is a deepfake, where images, audio or video resemble persons, objects, places or events in a way that would falsely appear authentic, and must disclose that text has been artificially generated or manipulated where it informs the public on matters of public interest, unless it has undergone human review and is subject to editorial responsibility. Anyone interacting with an AI system, a chatbot for instance, must be informed of that fact. All these obligations, as the Commission recalls, apply from 2 August 2026. With one exception, introduced by the Digital Omnibus: recital 38 provides for a four-month transitional period for providers who had already placed their systems on the market before 2 August 2026, which means that for those systems the marking obligation under Article 50(2) bites on 2 December 2026. The distinction must be kept in mind in any gap analysis, because it does not cover all transparency obligations, only marking, and it does not cover all providers, only those already on the market.
The code: two sections, a set of icons, an adequacy assessment
The code, published on 10 June 2026 following a multi-stakeholder process facilitated by the AI Office and entrusted to independent experts, is divided into two sections: the first addressed to providers, with the rules on marking and detection of generated or manipulated content; the second addressed to deployers, with the rules on labelling deepfakes and text. Alongside it, the EU has made available a set of icons that deployers may use to label AI-generated content, following the logic, familiar to data protection practitioners, of immediately intelligible visual communication. Signatories will be publicly listed during July 2026 and will take part in the Signatory Taskforces, which are to share practices and advance the implementation of marking and labelling. The architecture mirrors the one already tested for general-purpose AI models, which we examined when discussing GPAI models and the Code of Practice on Chapter V.
Adequate does not mean compliant, and the legislature now says so
The Commission states the legally most delicate point in spare terms: adherence to the code does not constitute conclusive evidence of compliance. Those who choose to comply by other means are not thereby in breach, but they will have to demonstrate that the measures adopted are adequate, and that demonstration will be assessed individually by the various market surveillance authorities. The practical benefit of signing lies elsewhere, namely in the fact that future enforcement will focus on monitoring adherence to the code, with reduced administrative burden and greater predictability regardless of the place of establishment and of the competent authority. So far, the Commission’s position. The Digital Omnibus adds a piece that points the same way and makes it explicit: in removing the Commission’s power to approve by implementing act the codes referred to in Article 50(7) and Article 56(6), the legislature reasons that those codes have limited legal effects and do not confer a presumption of conformity, so that approval by implementing act is not strictly necessary. This is precisely the distinction we encountered when reasoning about ISO/IEC 42001 and the presumption of conformity: an instrument may ease the burden of proof without producing the legal effect proper to a presumption, which under the AI Act remains reserved to harmonised standards. The code, then, is not a shortcut to compliance: it is a recognised map, and those who follow it will still have to show that they did.
Why high-risk slips and transparency, in substance, does not
The contrast with the rest of the Regulation is instructive. The Digital Omnibus, at recital 40, moves the application of Chapter III, Sections 1, 2 and 3 to 2 December 2027 for high-risk systems under Annex III and to 2 August 2028 for those embedded in products under Annex I, and it does so on openly stated grounds: the delay in the availability of technical standards, common specifications and guidance, and the delay in setting up national competent authorities, circumstances which, the text reads, undermine the effective start of application of those obligations and risk driving up compliance costs. We wrote about this when examining the new high-risk timeline. On transparency, by contrast, the supporting tools are there, the code, the icons, the guidelines on the scope of Article 50 now in preparation, and the date has not moved, save for the short transitional period granted to pre-existing systems. The regulation amending the AI Act will enter into force on the third day following its publication in the Official Journal of the European Union, which as at today has not yet taken place: the text referred to here is the one consolidated by the co-legislators, document PE-CONS 30/26.
In light of the foregoing, one may ask whether a code that the legislature itself describes as lacking full legal effects, and which for that very reason no longer warrants an implementing act, is destined to remain one option among others, or rather to become, in the practice of supervisory authorities and in the perception of the market, a de facto standard from which one departs only by assuming an evidentiary burden that few will be willing to bear.
