Today, 2 August 2026, the transparency obligations of Article 50 of the AI Act enter into application. After months of countdown, the date is no longer in the future. It deserves a brief note — precisely because, if the work is done, nothing spectacular happens today. And that's the point. It concerns anyone who provides or uses systems that converse with people or generate content.
What exactly changes today
From today, for every system Article 50 reaches — those already in use and those added later —:
- Systems that interact with people must make it known that they are AI — with the division of roles we already explained: the design is the provider's; the diligent verification, the deployer's.
- Synthetic content must be marked by its provider, and the deployer must disclose deep fakes and declare text published to inform the public — except via the editorial review route, which for most SMEs that publish with editorial judgement is the natural coverage.
- And as we reported this week, the market surveillance machinery that can ask about all of the above also starts up.
One transitional nuance worth keeping in mind: providers of generative systems placed on the market before today have until 2 December 2026 to comply with the Article 50(2) marking. It is neither an extension nor a general second chance: it is a transitional period created by Regulation (EU) 2026/1744, limited to those systems and their providers.
The figure allocates the four duties of Article 50: the notice that people are interacting with an AI system and the marking of synthetic content fall to the provider; the information on emotion recognition or biometric categorisation and the disclosure of deep fakes and of text published to inform the public on matters of public interest, to the deployer.
If you closed out July, today is Monday
The closing checklist we published a few days ago was exactly this: the sublist of affected systems, the chatbot check, the labelling criteria, the editorial workflow, and the record of all of it. Whoever completed it has no new task today: their notices were already visible yesterday, their criteria were already written, their file already carries a July date. A rule's entry into application is a non-event for those who prepared — which is the best possible description of good governance.
Those who didn't get there shouldn't tell themselves stories, but shouldn't panic either: the five tasks on that checklist are still the same, they still take days rather than months, and every day they're done and recorded is a day less of exposure. Starting late is worse than starting on time, and far better than not starting.
What comes next
The calendar doesn't stop here. On 2 December 2026 the marking transitional period expires for providers of systems that predate today, and the new Article 5 prohibitions on systems generating child sexual abuse material and non-consensual intimate images start applying, on the date set by Regulation (EU) 2026/1744. Beyond that, high risk under Annex III arrives on 2 December 2027 under the calendar set by the Digital Omnibus. Every date, like this one, will reward the same thing: having arrived with the evidence already built.
Today transparency stops being a plan. Welcome as it is: it was the most reasonable obligation in the Regulation, and the cheapest to get right.
Article 50 series → the four paragraphs on one page.
This article is for informational purposes only and does not constitute legal advice.