/
On 24 July 2026, the Official Journal of the European Union published Regulation (EU) 2026/1744, the so-called Digital Omnibus on AI. It amends Regulation (EU) 2024/1689 (the AI Act) — and, along the way, the aviation and machinery regulations — to simplify how the harmonised AI rules apply. The headline is the deferral of the high-risk obligations, but the detail matters: some duties are postponed and others still apply on their original date. Here is what actually changed.
What Regulation (EU) 2026/1744 is
It is a simplification («omnibus») regulation adopted by the European Parliament and the Council. It responds to a practical problem: delays in drafting the harmonised standards — which give providers the technical means to comply — and in setting up national governance and conformity-assessment frameworks had created a heavier burden than expected. The legislator answers by granting more time and clarifying application without lowering the level of protection of health, safety and fundamental rights that the AI Act pursues.
The key change: high-risk obligations are deferred
The AI Act timeline is re-ordered as follows:
- General application: 2 August 2026 — unchanged. Most of the regulation still starts on this date.
- Chapter III, sections 1, 2 and 3 (high-risk AI systems): 2 December 2027 — deferred.
- High-risk systems under Article 6(1): 2 August 2028 — deferred.
In plain terms: if your company develops or uses an AI system classified as high-risk, you now have more room to prepare for conformity. But «more room» is not «no longer applies»: the obligations are coming, and technical documentation, risk management and human oversight cannot be improvised at the last minute.
What does NOT change (and applies now)
This is the trap many companies will fall into. The omnibus does not postpone everything:
- Prohibited AI practices (Title II) have applied since February 2025.
- Rules for general-purpose AI (GPAI) models and governance have applied since August 2025.
- The transparency obligations of Article 50 — disclosing that content or an interaction is AI-generated — still apply from 2 August 2026. The omnibus only extends the deadline for the technical detail of marking synthetic content (Art. 50(2)), not the transparency duty itself.
Provider or deployer: your role defines your duties
The AI Act distinguishes two roles that are often confused. The provider develops the AI system (or markets it under its name); the deployer uses it under its authority in a professional context. Most SMEs — and most importers and manufacturers entering the EU — will be deployers of third-party AI tools. Deployer duties are lighter than provider duties, but not zero: for high-risk systems they include human oversight, use in line with the instructions, and record-keeping. Knowing which side you are on is the first step to neither over-complying nor missing an obligation that is genuinely yours.
Not just the AI Act: machinery and aviation too
The omnibus also amends Regulation (EU) 2023/1230 (machinery) and Regulation (EU) 2018/1139 (aviation) to set out how they interact with the AI rules and to avoid overlaps or inconsistent interpretation between sectoral and horizontal AI legislation. For a machinery manufacturer with AI components, this means the conformity assessment of the product and the AI Act assessment must be read together, in a coordinated way — not as two separate worlds. It is a recurring theme of the regulation: reduce friction between regimes without weakening any of them.
AI literacy: from strict duty to a duty to support
Article 4 (AI literacy of staff) is relaxed: it moves from a strict obligation to a duty to support and promote the training of the people operating the AI. It remains a strategic priority — a workforce that does not understand the tool is a risk — but the approach is now one of encouragement rather than immediate penalty.
Harmonised standards and presumption of conformity
The regulation reaffirms that harmonised standards cited in the Official Journal confer a presumption of conformity, and opens alternative routes (common specifications, guidance) while those standards arrive. This matters for EN 18286:2026, the European AI quality-management standard: it is already published, but not yet cited in the Official Journal, so it does not yet grant an automatic presumption of conformity. Aligning with it today is good practice; presenting it as «a standard in force that grants presumption» would be premature.
What it means for your company — a checklist
- Classify your AI systems by risk level (prohibited, high, limited, minimal). Everything starts here.
- If you have high-risk AI (for example, AI-enabled medical device software), use the new 2027-2028 timeline to prepare technical documentation, risk management and human oversight — not to forget about it.
- Comply with Article 50 transparency now if you generate content with AI or deploy chatbots: it applies from August 2026.
- Train your team on the responsible use of AI (Art. 4).
- Document who reviews, validates and signs off each AI output with regulatory impact.
The bottom line
Regulation (EU) 2026/1744 buys time where the market needed it most — high-risk conformity, where the standards were not ready — while keeping the parts that protect people (prohibitions, transparency, governance) firmly in force. For most companies the right reading is not «the AI Act is delayed, relax», but «we now have a realistic runway to do this properly». The organisations that use the extra eighteen-plus months to classify their systems, build the documentation and put real human oversight in place will arrive at 2027-2028 ready; those that read it as a reprieve will arrive late.
Frequently asked questions
Has the AI Act been postponed entirely? No. Only the high-risk obligations (Chapter III and Art. 6(1)) move to 2027-2028. General application stays on 2 August 2026, and prohibitions, GPAI and transparency already apply.
Can I stop worrying about the AI Act until 2027? No. Transparency applies now, and preparing high-risk conformity takes months: the extra time is for working, not for waiting.
What about AI-enabled medical devices (SaMD)? AI medical software is both a medical device (MDR/IVDR) and, depending on its function, a high-risk AI system. Under the omnibus the AI Act high-risk layer aligns with the 2027-2028 timeline, but MDR/IVDR obligations run on their own schedule: read both together, do not pick one.
Who is liable if the AI gets a file wrong? In an operated service like ASC’s, the specialist who validates and signs the output answers for it — not a disclaimer that shifts the risk onto the client. That is the difference between an accountable service and a self-service tool.
How we handle it at ASC
At ASC we help companies bring order to their use of AI within the regulatory framework: we classify systems by risk, prepare the documentation and the Article 50 transparency duties, and — most importantly — we operate a model where AI produces and a qualified specialist validates and signs off every output. We do not sell a tool for you to use on your own: we provide an operated service with accountability. If the regulator stops your file, we answer — the file carries our specialist’s signature, not a disclaimer that shifts the risk to you. If your company uses AI in a regulated context, let’s talk through our contact page; see also our note on EN 18286:2026 and on regulatory affairs automation. Official sources: Regulation (EU) 2026/1744 and Regulation (EU) 2024/1689 (AI Act).