Corrected on 9 October 2026
We wrote that from 2 August all generated text, images and video must carry a machine-readable mark. Regulation (EU) 2026/1744 of 8 July 2026 added Article 111(4) to the AI Act: systems generating audio, images, video or text that were placed on the market before 2 August 2026 have until 2 December 2026 to meet the marking rule in Article 50(2). Article 50 and the 2 August date for new systems are unchanged. The dek, the brief and the fact box were corrected and the consolidated text on EUR-Lex was added. The conclusion doesn't change.
From 2 August, customer-facing AI stops being a matter of taste and budget. It becomes a matter of rule - with a date, and with something you have to show.
Fact
The facts: on 20 July 2026 the European Commission published guidelines on applying the transparency obligations under Article 50 of the AI Act. The obligations themselves start applying from 2 August 2026. Under them, providers must design systems so a person is notified when interacting directly with AI, and add a machine-readable mark that allows AI-generated or altered content to be recognized. There is one exception for the mark: systems generating audio, images, video or text that were placed on the market before 2 August 2026 must comply with Article 50(2) by 2 December 2026 (Article 111(4), added by Regulation (EU) 2026/1744 of 8 July 2026). Deployers - those who put the system in front of people - must disclose when a person is exposed to a deepfake, to AI-generated content on matters of public interest without human editing, or to emotion recognition or biometric categorization systems. Alongside the guidelines, the Commission also prepared a Code of Practice on transparency for AI-generated content, covering marking and labeling. Source: European Commission, Shaping Europe's digital future, 20.07.2026; EUR-Lex, consolidated text of the AI Act as of 27.07.2026.
In practice it splits in two. If you build the system - it's your job to put the mark inside, and to have the chatbot say for itself that it's a chatbot. If you just use it in front of customers - then you disclose: that the other side is a machine, that the video is a deepfake, that the text was written by a model with no human review. The small business here that puts a 'virtual assistant' on the site or runs a generated ad falls right into the second bucket.
The machine has the right to speak. It has no right to pretend to be human.
The guidelines themselves aren't a new law. They explain how to read the article, so it isn't interpreted a different way by everyone. What's firm is the other part: the date. By 2 August it's wise to go through every place where AI touches the customer - chat, content, ads - and put the label on, before someone asks you for it.
This isn't a brake. It's a direction we hold to anyway: say what's machine. A customer who finds out on their own that the 'person' in the chat was a bot walks away. A customer you told upfront stays. The rule just writes into law something good practice already does - not hiding who's talking.
Sources
European Commission - Commission publishes guidelines on transparency obligations for Article 50 (news), Shaping Europe's digital future, 20.07.2026European Commission - Guidelines on Article 50 of the AI Act (library)EUR-Lex - AI Act, consolidated text as of 27.07.2026 (Article 111(4), added by Regulation (EU) 2026/1744)The visual is generated code-art, no third-party images.