The EU AI Act high-risk delay is a gift. Don't waste it

First published in Norwegian on Digi, 1 April 2026

High-risk AI system requirements will likely be pushed back. Most organizations are breathing a sigh of relief and slowing down. My advice: do the opposite!

Members of the European Parliament have voted to delay the application of requirements on high-risk AI systems, to give authorities more time to prepare harmonized standards that will help organizations actually comply with these requirements.

The Parliament and the European Commission are aligned on the delay, though it still requires approval from a third institution, the Council of the European Union, before it is confirmed.

On Linkedin, it’s clear that industry response is relief. I’ve had collaborators ask if we should cancel our planned high-risk requirements training.

I completely understand the relief!

But to my clients and others who have been working hard to meet august 2026 deadlines: now is your chance to keep going and show market leadership.

Continuing your AI Act compliance work will signal to your own users / customers / citizens:

1️⃣ You are committed to protecting safety and fundamental rights.

2️⃣ You use/build AI you're comfortable defending publicly, before you are legally required to.

3️⃣ You are the kind of organization that national monitoring bodies want to work with, not pursue.

4️⃣ They have a reason to trust you.

𝐀𝐧𝐝 𝐲𝐨𝐮 𝐡𝐚𝐯𝐞 𝐭𝐡𝐞 𝐨𝐩𝐩𝐨𝐫𝐭𝐮𝐧𝐢𝐭𝐲 𝐭𝐨 𝐛𝐞 𝐟𝐢𝐫𝐬𝐭.

Full disclosure: I sit on the European Joint Technical Committee (JTC 21) working directly on the harmonized standards the European Commission has ordered to support AI Act implementation. Through Standard Norway, I also lead the working group responsible for channeling Norwegian input into standards covering risk management, quality management systems, and the evaluation of bias in AI. In other words, I am one of the people helping to write the very guidance this delay is meant to create space for.

So when I say the delay is not a reprieve - it is a window - I mean it with full knowledge of what is coming.

High-risk AI requirements haven’t changed, only the timeline has

From where I sit in the standards process, I can tell you directly: the direction of travel is not changing. The harmonized standards we are developing are designed to make compliance clearer, not lighter.

The organizations that wait until 2027, won't have an easy time. These standards will not help if you have already developed or implemented a high-risk system without thinking about human oversight, fundamental rights, or explainability.

Compliance done under pressure looks like compliance. Compliance pursued by choice looks like leadership.

Use this time to do the work (better)

I am not suggesting you ignore the practical realities of the delay. By all means, use the extra time well.

-- to stress-test your risk classification decisions.

-- to strengthen your human oversight mechanisms.

-- socialize understanding of why these requirements exist (to protect fundamental rights)

-- 1000% for those using agentic AI - to make sure you are building all of the above in from the beginning.

-- to get comfortable with the emerging harmonized standards - you will have a huge head-start. Even if full certification isn't on the table for you, these standards will give high-quality structures to align with.

A chance for market leadership

Full AI Act enforcement will come. The harmonized standards will land. When they do, some organizations will be scrambling to comply at the last minute, while those that used the entire delay window will be able to easily differentiate themselves.

About me, Ley Muller: I founded Values-driven AI to work with scale-ups, SMEs, and public institutions across Norway on responsible AI and AI Act readiness — and I bring direct insight from the European standards process to every engagement. Please feel free to reach out if you are interested in training, leadership workshops, advising, or help articulating your use or sales of AI, into market advantage.