What changed before the transparency rules began to apply?
Article 50 of Regulation (EU) 2024/1689 sets transparency duties for certain AI interactions and synthetic or manipulated content. The base AI Act was adopted and published in 2024 and entered into force on 1 August 2024. Its staggered application rule makes Article 50 generally applicable from 2 August 2026.
The legal position changed again shortly before that date. Regulation (EU) 2026/1744 was adopted on 8 July 2026, published in the Official Journal on 24 July and entered into force on 27 July, the third day after publication. Article 1, point 20 of that amending Regulation replaced Article 50(7). It also created a specific transition: providers of generative AI systems placed on the market before 2 August 2026 must comply with Article 50(2) by 2 December 2026.
That four-month transition is narrow. It concerns the paragraph 2 machine-readable marking duty for the specified pre-2-August generative systems. It is not a general postponement of Article 50, and it should not be used to shift every provider or deployer duty to December. For a live compliance decision, check the current EUR-Lex text and the facts of the particular system. This guide is general information, not individual legal advice.
First separate providers from deployers
A provider develops an AI system or has it developed and places it on the market or puts it into service under its name. A deployer uses an AI system under its authority, subject to the Act’s definitions and exceptions. One organisation may occupy different roles across different products. A company offering an interactive system may have provider responsibilities, while a publisher using that system in a public-facing workflow may have deployer responsibilities.
The distinction matters because the duties are not interchangeable. Providers of systems intended for direct interaction with natural persons must inform those persons that they are interacting with AI, unless this is obvious to a reasonably well-informed, observant and circumspect person, taking into account the circumstances and context. The direct-interaction duty has an exception for systems authorised by law to detect, prevent, investigate or prosecute criminal offences, subject to appropriate safeguards for third-party rights and freedoms; that exception does not cover systems available for the public to report a criminal offence.
Article 50(2) addresses providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text content. Outputs must be marked in a machine-readable format and be detectable as artificially generated or manipulated. The measures must be effective, interoperable, robust and reliable as far as technically feasible, taking account of the specificities and limitations of different content types, implementation costs and the generally acknowledged state of the art. The duty does not apply where the system performs a standard editing function or does not substantially alter the input data supplied by the deployer or its semantics. It also contains the authorised criminal-law exception, subject to safeguards.
Deployers have separate duties. A deployer of an emotion-recognition or biometric-categorisation system must inform exposed natural persons and process personal data in accordance with the applicable data-protection rules, subject to the authorised criminal-law exception. A deployer that generates or manipulates image, audio or video constituting a deepfake must disclose that the content was artificially generated or manipulated. A deployer publishing AI-generated or manipulated text to inform the public on matters of public interest must also disclose that fact, subject to the exact exceptions described below.
For the disclosures covered by Article 50(1) to (4), the information must be clear and distinguishable and provided no later than the first interaction or exposure. It must also comply with applicable accessibility requirements. A buried statement after the interaction is therefore not the same as a timely disclosure.
Visible disclosure and machine-readable marking are different controls
A visible notice helps a person understand the immediate experience. A machine-readable mark supports technical detection and downstream identification. Depending on the workflow, both controls may matter; neither should automatically be treated as a substitute for the other.
Product teams should test placement, timing, accessibility, translated interfaces and small-screen behaviour. Engineering teams should test whether marking survives realistic export and transformation paths. Governance teams should retain versioned evidence showing which measure operated in the released system, for which role and use case, and under which exception or transition analysis.
Keep the deepfake and public-interest exceptions exact
The deepfake disclosure duty does not apply to authorised criminal-law uses. Where the content forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme, the obligation is limited to disclosure of the existence of generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work.
For AI-generated or manipulated text published to inform the public on matters of public interest, disclosure likewise does not apply to authorised criminal-law uses. It also does not apply where the content has undergone a process of human review or editorial control and a natural or legal person holds editorial responsibility for publication. Those conditions should be documented honestly: attaching a person’s name to an otherwise automated pipeline does not by itself prove meaningful review or editorial control.
A practical implementation sequence
- Inventory direct interactions. List chat, voice, support and other interfaces where a person may reasonably believe they are dealing with a human.
- Map legal roles. Record the provider and deployer for each use case rather than assigning every duty to “the vendor.”
- Classify outputs. Identify synthetic audio, image, video and text flows, including deepfakes and public-interest publication.
- Check timing and transition. Separate the general 2 August 2026 application date from the paragraph 2 transition ending 2 December 2026 for qualifying systems already on the market.
- Design human and technical controls. Specify the disclosure, machine-readable marking, timing, language and accessibility behaviour that applies to the classified workflow.
- Test real journeys. Verify first contact, copied or exported content, mobile layouts, assistive technology and failure states.
- Keep decision evidence. Preserve versions, role maps, exception analyses, review records and test results so the organisation can explain its approach.
Guidelines and codes do not replace the amended law
Commission guidelines and a transparency code of practice can be useful implementation aids, but they should be read against the current enacted text. Regulation (EU) 2026/1744 replaced Article 50(7); a service-desk page, proposal-era FAQ or stale-text banner cannot override the Official Journal act. Nor should participation in a voluntary code be described as a blanket safe harbour for every Article 50 duty. Teams should identify which statutory paragraph, role and content flow a given measure addresses, and verify the current scope before relying on it.
The reader’s practical takeaway
Article 50 is not solved by placing “powered by AI” in a footer. A credible programme links legal classification, product design, technical provenance, editorial practice and evidence. Readers should receive information early enough to calibrate trust, while organisations should be able to show that a notice or mark remains effective when the model, interface, audience or distribution path changes.
As of 11 September 2026, this is a practical overview of Article 50 and its narrow transition. It is not a complete analysis of Regulation (EU) 2026/1744 or of the amended AI Act, and specific deployments require review of the current text and qualified advice.
Primary sources
EUR-Lex: Regulation (EU) 2026/1744, CELEX 32026R1744; EUR-Lex: Regulation (EU) 2026/1744, document information and versions; EUR-Lex: Regulation (EU) 2026/1744, EN–ES text view; EUR-Lex: Regulation (EU) 2024/1689, document information and modification record; European Commission: guidelines on AI transparency obligations. Accessed 11 September 2026.