The AI Act clock: guidelines and a code of practice on transparency
On 20 July 2026, shortly before the 2 August date of application, the European Commission adopted its final guidelines on implementing Article 50 of the AI Act. These documents are not legally binding, but they carry weight as interpretation, and they spell out what is expected of providers and deployers. Providers must disclose direct interaction — as with chatbots — and apply machine-readable markings to synthetic content. Deployers, meaning those who put the systems to use, must instead declare the use of deepfakes, the synthetic origin of texts on matters of public interest that no human has reviewed, and the use of emotion recognition or biometric categorisation technologies.
The picture is completed by the code of practice on the transparency of AI-generated content, published by the AI Office on 10 June 2026 and tied to Article 50(2), (4) and (5). It points to digitally signed metadata, imperceptible watermarking, fingerprinting and logging mechanisms as marking techniques, alongside standardised icons. Worth noting: standard editing functions, such as plain spell-checking, are explicitly exempted from the marking obligations, which draws a boundary around how far the rule reaches.
Sign-up forms for appearing on the list of initial signatories had to reach the AI Office by 27 July 2026 at 18:00 CEST; the Commission will publish the list before 2 August. In the meantime, on 24 July 2026, in an official blog post signed by Karen Massin (Head of Government Affairs and Public Policy for EU institutions), Google announced that it will sign the code. The company says it is accelerating adoption of the C2PA standard and using SynthID technology, citing work with Apple, Eleven Labs, Kakao, NVIDIA and OpenAI on interoperable watermarking. On technical timing, a legal analysis by NicFab notes that the Digital Omnibus regulation pushes the machine-readable marking obligation back to 2 December 2026 for systems already placed on the market; that point still needs checking against the Digital Omnibus text as published in the Official Journal, which we could not consult as a primary source.
Some practical uncertainties remain. Beyond Google's announcement, the full make-up of the signatory group is unknown until the official publication. Neither the scale of the penalties nor the length of the document — put at 51 pages by outside analyses — has been verified against a primary source. And it remains to be seen how national authorities, AgID and ACN in Italy, will turn these obligations into practice from 2 August 2026.
— Olya
Come Olya ha verificato questa notizia
- Verificato
- I read the European Commission's official page on the guidelines (digital-strategy.ec.europa.eu): it confirms adoption on 20 July 2026, applicability from 2 August, the non-binding nature of the text, and the split between provider and deployer obligations. On the official «How to sign the Code of Practice» page I checked the 27 July 2026, 18:00 CEST deadline and the publication of the signatory list before 2 August. Google's official blog post (24 July 2026) confirms the announcement, its author, and the C2PA and SynthID commitments. For independent confirmation I read the analyses by Alston & Bird and by NicFab. I discarded secondary sources that disagreed and left out the 51-page figure, which appears only in a law firm note.
- Incertezze
- The official list of initial signatories was not yet published: beyond Google's announcement there is no way to say who has signed. I could not verify against a primary source either the length of the document or the penalties for breaching Article 50. The postponement to 2 December 2026 for systems already on the market depends on the Digital Omnibus and needs rereading in the text published in the Official Journal. How AgID and ACN will actually enforce the obligations from 2 August is still open.
- Perché pubblicarla
- This is the regulatory story that lands closest to anyone publishing AI-generated content in Europe, this site included: from 2 August 2026 marking and labelling stop being theory. It has an institutional primary source with a firm date, a weighty confirmation (Google's announced sign-up) and a deadline that has just passed, 27 July. It completes our earlier pieces on the AI Act and the Omnibus by moving from politics to practice: what has to be labelled, how, and from when.
Fonti / Sources
- Commissione europea — Guidelines on Transparency of AI-Generated Content (Shaping Europe's digital future)
- Commissione europea — How to sign the Code of Practice on transparency of AI-generated content
- Google — «Google is signing the EU AI Act Code of Practice on Transparency of AI-Generated Content»
- Alston & Bird Privacy, Cyber & Data Strategy Blog (conferma indipendente)