By Florin Cusmereanu · AI Fluent · July 2026 · 7 min read Category: EU AI Act & Governance
The Digital Omnibus Moved the EU AI Act Deadlines. What That Means for Manufacturers
For most of the past year, the compliance conversation in manufacturing has been organised around a single date: 2 August 2026. That date carried the high-risk obligations, and it is the reason a lot of plants opened an AI governance file in the first place.
That date has moved. The Digital Omnibus package, adopted by the European Parliament and the Council in June 2026, pushes the high-risk requirements out by well over a year. The headline that followed in most trade press was "the AI Act is delayed", which is close enough to be repeated and wrong enough to cost you.
Here is what actually changed, what did not, and what a manufacturer should do with the extra runway.
What the Digital Omnibus is
The Digital Omnibus is a targeted amendment package to existing EU digital legislation, including the AI Act. The Council announced political agreement and adoption in its press release 553/26 of 29 June 2026, and the agreed text is carried in PE-CONS 30/26.
Publication in the Official Journal is still pending. Until that publication is confirmed, treat any precise entry-into-force date you see quoted as provisional and verify it at EUR-Lex before it goes into a policy document or a board paper. The substance below is settled. The exact commencement day is the part still in transit.
What changed
The high-risk dates moved out. Obligations for high-risk AI systems classified under Article 6(2) and Annex III now apply from 2 December 2027. High-risk AI embedded in regulated products under Article 6(1) and Annex I applies from 2 August 2028. For a manufacturer looking at AI in quality inspection, in workforce decision support, or embedded in CE-marked machinery, that is the practical planning horizon now.
Worth being precise about the scope of the delay, because this is where the trade coverage went loose. The postponement covers Chapter III Sections 1 to 3, with Article 6(5) excepted. It is not a blanket suspension of everything labelled high-risk, and it is not a suspension of the Act.
Article 4 was softened in wording, not in force. The AI literacy provision now reads as an obligation to take measures to support the development of AI literacy, where the earlier text required organisations to ensure a sufficient level. That is a change in how the duty is expressed. It is not a delay and not a removal. Article 4 has applied since 2 February 2025 and continues to apply.
Two things arrive on 2 December 2026. New prohibitions enter application under Article 5, covering non-consensual intimate imagery and the generation of child sexual abuse material. Separately, providers of generative AI systems that were placed on the market before 2 August 2026 have to meet the Article 50(2) transparency requirement by that date, which means machine-readable marking of synthetic content. If your organisation has shipped or embedded a generative feature in a product already on the market, that retrofit is the nearest real deadline in front of you.
What did not change
AI literacy is a live obligation. It has been since February 2025. Softer wording does not create a grace period, and a supervisory authority asking what measures you took will not accept "the deadline moved" as an answer, because this deadline did not.
The prohibitions are in force. Practices banned under Article 5 were banned before the omnibus and remain banned.
Risk classification is still the first step. Nothing in the amendment changes the fact that you cannot manage an obligation you have not identified. Knowing which AI systems you operate, in which contexts, and in which tier is the work that everything else depends on. It was step one before June 2026 and it is step one now.
Penalties are unchanged. The enforcement regime was not touched.
The honest way to read the package: you were given runway, not a repeal. The obligations arrive later, in the same shape, at organisations that will have had 18 additional months to prepare. Regulators will know that.
What to do with the extra time
The organisations that used the original deadline well were not the ones that moved fastest. They were the ones that knew what they had. That is still the highest-value use of the next 12 months.
Finish the inventory. List every AI system in operational use, including AI features inside software you bought for another purpose. Copilot in the purchasing inbox counts. An AI-assisted vision check on a line counts. Assign each one a use context and a probable tier. This is a half-day exercise with operations, IT, HR and purchasing in the room, and most plants find one or two systems they had not thought of as AI at all.
Close the Article 4 gap now. A written AI usage policy that names the tools, the contexts and the expectations, plus documented role-specific training for the people using those tools. Neither requires a certification programme. Both take longer to assemble than anyone estimates, which is the argument for starting while the calendar is generous.
Use the runway on the systems that will land in 2027 and 2028. Technical documentation, data governance, human oversight design and conformity assessment are slow work. A system you plan to deploy in 2027 is better designed against those requirements now than retrofitted against them later. The cost difference between the two is not small.
Check the 2 December 2026 items against your own products. Most manufacturers will find nothing here. If you have a generative feature in a product already on the market, you have a specific and near-term piece of work.
Where to start
The ACTS assessment (AI Compliance Tier Score) gives you an initial read on your exposure in under 5 minutes, against the current dates rather than the ones the trade press is still quoting. No email required to see your tier. It is a starting point, not a legal opinion.
If the result raises questions worth working through with someone who has run governance inside manufacturing operations and knows the TISAX and ISO 27001 overlap:
Book a governance conversation → 30 minutes. We look at the AI systems you are operating and give you a clear picture of your actual exposure. No prep needed.
For the full plain-English guide to the Act, the risk tiers and the current timeline:
Related reading:
- Why the Back Office Is Where Industrial AI Should Start → (Industrial AI)
- The First Automation You Should Build → (SMB Automation)
Share this article: [LinkedIn →]
AI Fluent is not a law firm. This article is provided for general awareness and does not constitute legal advice. Compliance responsibility rests with your organisation. For legal interpretation of specific obligations under the EU AI Act, consult qualified legal counsel.