The EU AI Act just changed in 2026, and what it means for AEC
Anyone who spent last year bracing for a hard EU AI Act deadline on 2 August 2026 can take a breath. The date came and went, and the obligation most firms feared did not switch on. In late July 2026the EU brought in a package of amendments, the Digital Omnibus on AI, that pushed the heaviest duties back by more than a year. For construction, engineering and architecture firms, the honest read on the EU AI Act in 2026 is that the pressure eased on one front and stayed exactly where it was on another. This piece sets out what moved, what did not, and what a built environment firm should actually do about it.
What actually changed in 2026?
The core of the change is a delay to the high-risk rules. Obligations for standalone high-risk AI systems, the ones listed in Annex III of the Act, have moved from 2 August 2026 to 2 December 2027. For AI built into products already covered by EU product safety law, Annex I, the date has moved to August 2028. The reason given was practical rather than political: the technical standards and the assessment bodies the Act relies on were not ready in time, so the timeline was adjusted to match reality.
What did not move is just as important. The transparency rules under Article 50, which cover telling people when they are dealing with an AI system and labelling AI-generated content, still began on 2August 2026. And the two duties that have applied since February 2025, the AI literacy obligation and the ban on prohibited practices, are untouched. They were in force before this change and they remain in force now.
So the picture is a delay on the complex, higher-risk end and no change at all on the everyday duties that most firms actually fall under.
Does the delay let AEC firms off the hook?
For most construction and engineering work, the delay was never the main event. The high-risk category covers uses such as recruitment and worker management, credit and access to essential services, law enforcement and border control. A quantity surveyor drafting a cost report with Copilot, an engineer summarising a site report, or an architect using AI to speed up a tender response is almost always in the limited-risk band, not the high-risk one.
That matters because the duty that lands on limited-risk use is the AI literacy obligation, and that has been live since February 2025. Every organisation using AI has to make sure the people using it have a sufficient understanding of what these tools do, where they fail, and how to use them responsibly. There is no delay on that. Understanding the detail of EU AI Act compliance is the starting point, and it applies to a two-person practice as much as a national consultancy.
There is one place the delay is worth watching. If a firm uses AI in hiring or to manage and assess its own people, that sits in the high-risk category, and the December 2027 date is the one to hold in mind. For most day-to-day design and delivery work, though, the literacy duty is the here-and-now obligation.
Why the delay is a reason to move now, not later
A postponed deadline reads like breathing room, and it is easy to file AI compliance under "later". The firms getting real value from AI are reading it the other way around. The delay buys time to build capability properly instead of scrambling to meet a date, and the everyday duty to train people is already active regardless.
Two things tend to happen inside a firm over the next eighteen months. AI use spreads, often faster than anyone plans, as more people find Copilot and Claude useful on real work. And clients, insurers and professional bodies start asking how a firm governs that use. A practice that has trained its people and written down how it works is ready for both. A practice that waited for a deadline is not. Firms across the built environment that treat this as capability building rather than a compliance scramble are the ones reporting around 22% productivity gains and more than four hours saved per person each week.
What should a construction or engineering firm do about it?
The action list has not really changed with the new dates. It is short and practical:
- Train your people to the literacy standard. This is the live duty. Role-based training, matched to what an engineer, a QS or an admin lead actually does, is how a firm meets it and gets the productivity at the same time.
- Get transparency right. Where AI writes or generates client-facing content, be clear about it and label AI-generated material where the Article 50 rules apply.
- Write down how you use AI. A short inventory of approved tools, clear data rules, and named human checks covers most of the near-term duty. A working AI governance approach for construction firms turns scattered use into managed practice.
- Flag any high-risk use early. If AI touches hiring or worker assessment, note it now and plan for the December 2027 date rather than being surprised by it.
The thread running through all four is capability. The law asks for people who understand the tools and a firm that can show how it manages them. Both come from training, not from a policy nobody reads.
How this connects to AEC training
Governance sets the rules and training makes them real. A rule a person cannot map to their own task changes very little, which is why the strongest results come when the compliant way of working and the productive way are taught as the same thing. When a surveyor learns Copilot on a live cost report and sees where the safe line for client data sits, the rule and the skill land in the same moment. That is the logic behind EU AI Act compliant training and AI literacy training built for the built environment: the everyday duty and the dailyvalue arrive together.
FAQ
Did the EU AI Act get cancelled or paused in 2026?
No. The Act is still law. The Digital Omnibus on AI delayed the high-risk obligations, moving standalone high-risk systems to December 2027 and product-embedded AI to August 2028. The AI literacy duty and the transparency rules were not delayed.
Does the AI literacy duty still apply to construction firms?
Yes. It has applied since February 2025 and it covers any organisation using AI, whatever its size. Most everyday AEC use of AI falls under this duty rather than the high-risk rules.
Is our use of Copilot or Claude high-risk under the Act?
For typical design, drafting and admin work, no. That sits in the limited-risk band. High-risk covers uses such as recruitment, worker management and access to essential services. This is guidance, not legal advice, so check specific uses against the current rules.
If the deadline moved, why train people now?
Because the literacy duty is already live, AI use is spreading inside firms regardless, and clients and insurers are starting to ask how you govern it. Building capability now is easier than meeting a deadline later.
Where to start
Start with your people. Give each role the training the AI literacy duty expects, get transparency right where AI writes client-facing content, and write down how the firm uses AI. That covers most of the current obligation and turns a moving regulatory picture into steady, productive practice. The dates will keep shifting. A trained, well-governed firm stays ready whichever way they move.
Build capability and compliance with us ,or see the courses built for the built environment.
