The European Commission published its final guidelines on the AI Act’s transparency obligations on 20 July 2026, less than two weeks before those obligations begin to apply on 2 August. The 51-page document tells providers and deployers of AI systems how to meet Article 50 of the Act, which requires people to be told when they are interacting with AI and when content has been generated or altered by it, with the stated aim of reducing deception and manipulation. We covered the Act’s transparency rules when it entered into force. What is new is that the abstract obligation now has official implementation guidance, a confirmed Code of Practice, and a start date measured in days.
What the guidelines say
The practical detail matters more than the framework. Labelling of AI-generated content must sit on the content itself, not in metadata or in a caption that can be cropped away. The guidance offers concrete examples, and one of them is an email: a message generated by an AI agent to a natural person should carry an AI notice at the top. Chatbots and AI agents must disclose their artificial nature at the point of interaction, through written notices, persistent badges, or audio cues, with no single format prescribed. Providers of generative AI systems must build reliable, machine-readable markings into generated content so that synthetic material can be detected as such. The Commission has also formally assessed the voluntary Code of Practice on marking and labelling AI-generated content, published in June, as an adequate compliance route.
The guidelines are non-binding, and only the Court of Justice can interpret the Act authoritatively. In practice, national market surveillance authorities and the AI Office can be expected to follow them.
What actually binds email and martech teams
The scope is narrower than the headlines suggest, and worth stating precisely. The text-labelling obligation in Article 50(4) covers AI-generated text published with the purpose of informing the public on matters of public interest. Routine marketing copy drafted with AI assistance is generally not that, and there is an exemption where AI-generated text has undergone human review and a person holds editorial responsibility. So the obligation does not turn every AI-assisted subject line into a compliance event.
Three things do land on this industry. First, interactive systems: customer-service chatbots and AI agents that converse with customers, including agents that send email, must disclose themselves from 2 August, and the guidance’s email example makes the application explicit. Second, publications: newsletters and content programmes that function as public-interest publications sit closer to the labelling line than campaign mail does, and the cut-off for text is the date of publication, so AI-generated text published on or after 2 August needs labelling even if it was written earlier, unless the editorial-control exemption applies. Third, the provider question: ESPs and martech platforms that embed generative AI are potentially providers under the Act, carrying the machine-readable marking duty for content their systems generate. That is a product obligation, not a marketing one, and vendors should know which side of the provider-deployer line their features fall on.
The deadline confusion, resolved
Amid wide reporting that the AI Act is being delayed, teams should be precise about what has and has not moved. The Digital Omnibus simplification package, adopted by the Council in June and awaiting publication in the Official Journal, defers one thing: the machine-readable marking conformity deadline for generative systems already on the market before 2 August, which moves to 2 December 2026. Everything else in Article 50, including chatbot and agent disclosure, deepfake labelling, and public-interest text labelling, applies from 2 August without exception. Penalties for transparency breaches run to €15 million or 3 per cent of worldwide annual turnover, whichever is higher. This lands in the same season as the CNIL and Garante pixel rules, a European regulatory run we mapped in our analysis of the spring’s overlapping measures, and alongside the AI wiretap class actions testing adjacent theories in the United States.
What to do before 2 August
Audit where AI-generated text and AI interaction enter customer communications now, not in August. Map every chatbot, agent, and automated flow that converses with customers, and confirm each discloses its artificial nature at first contact. Classify content programmes honestly: campaign mail with human editorial control on one side, public-interest publishing on the other, and document the review process that supports the exemption where you rely on it. Ask your ESP and martech vendors where they sit on the provider-deployer line and how their generative features handle machine-readable marking, including the December conformity date for pre-existing systems. Legal, compliance, and marketing operations should agree which use cases require disclosure, labelling, or human review before volume scales, because after 2 August the question is no longer best practice.
More info: European Commission announcement of the transparency guidelines, 20 July 2026 and the guidelines library page. Additional analysis: Bird & Bird’s first impressions of the final guidelines.
