The omnibus regulation (EU) 2026/1744 postpones the obligations on high-risk AI systems: August 2028 for AI embedded in medical devices. But 2 August 2026 remains a real deadline: transparency, authorities and penalties enter into application. What it means for SaMD manufacturers.
The AI Act (Regulation (EU) 2024/1689) was due to apply in almost its entirety on 2 August 2026. The omnibus regulation on AI (Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026) has just changed this timeline. The obligations on high-risk AI systems have been postponed, until as late as 2 August 2028 for AI embedded in medical devices.
This postponement does not strip the 2 August 2026 deadline of its substance. The transparency obligations, the surveillance authorities and the penalty regime do enter into application on that date. Here is what applies, what has been postponed, and what it changes in practice for a Digital Medical Device (SaMD) manufacturer.
The AI Act's initial timeline set general application for 2 August 2026, followed by entry into application on 2 August 2027 for high-risk AI systems falling under Annex I, that is, AI that is a product, or a safety component of a product, already covered by EU harmonisation legislation such as the MDR (Medical Device Regulation, Regulation (EU) 2017/745) or the IVDR (In Vitro Diagnostic Regulation, Regulation (EU) 2017/746).
Approved by the European Parliament on 16 June 2026, Regulation (EU) 2026/1744 shifts both deadlines:
The legislator is open about the reason for the delay: the harmonised standards, guidance and support tools needed for compliance are not ready. The spirit of the text is to shift the obligation without shifting the preparation work.
The postponement covers only the obligations on high-risk systems. The rest of the regulation follows its initial timeline.
What is already in force. Since 2 February 2025, Article 5 has prohibited certain uses of AI considered incompatible with fundamental rights, while Article 4 requires organisations to ensure that the people who develop, deploy or use AI systems have a sufficient level of knowledge and skills to master them. Since 2 August 2025, providers of general purpose AI models (GPAI) must, among other things, document their models, inform the actors who integrate them into their own systems and, for models presenting a systemic risk, assess and mitigate those risks.
Article 50 transparency. From 2 August 2026, any AI system that interacts directly with people must disclose it, AI-generated or manipulated content must be marked, and users must be informed in cases of emotion recognition or biometric categorisation. A digital health company quickly falls within scope: a patient-facing chatbot, automatic generation of reports or letters, avatars or synthetic content in an application. These obligations have not been postponed.
Authorities and penalties. National market surveillance authorities are operational on 2 August 2026 and the penalty regime applies: up to 35 million euros or 7% of worldwide turnover for prohibited practices, and up to 15 million euros or 3% for breaches of the other obligations, including Article 50 (Article 99).
1. Place your system in the right timeline. A medical device embedding AI and going through an NB falls under Annex I, with a deadline of 2 August 2028. Article 50 transparency features apply from 2 August 2026, with no postponement. A single product can combine several of these situations, and your regulatory scoping must tell them apart.
2. Treat Article 50 as an immediate deadline. Inventory the points of contact between your AI and its users: direct dialogue, generated content, voice synthesis. Each case calls for clear user information or content marking. It is a light workstream compared with the high-risk requirements, but it is enforceable now and subject to penalties.
3. Use the twelve months gained to work through the AI Act / MDR interplay. The high-risk requirements (risk management, data governance, technical documentation, human oversight, post-market surveillance) largely overlap with what the MDR already requires. The AI Act allows the AI documentation to be integrated into the existing Technical Documentation and assessed as a whole by the NB. Mapping these overlaps now avoids paying twice for the same work.
4. Do not freeze the workstream. The postponement is precisely meant to leave time to prepare, not to defer preparation. Harmonised standards will arrive by 2028, NB capacity will remain limited, and the substantive requirements themselves have not changed. A manufacturer that builds the AI requirements into its Quality Management System (QMS) as development progresses will reach 2028 with a file already in place.
Regulation (EU) 2026/1744 postpones the obligations on high-risk AI systems to 2 August 2028 for AI embedded in medical devices. 2 August 2026 remains a real deadline: Article 50 transparency, the surveillance authorities and the penalties enter into application. For a SaMD manufacturer, the course of action is clear: address transparency without delay, place each system in the right timeline, and use the time gained to align AI requirements with the MDR file rather than to suspend the work.
Sparta Care supports healthcare software and AI system companies in their compliance journey:
Contact us to place your AI systems in the right timeline and turn the postponement into a head start.
