Europe Has an AI Act and Is Waiting for a Socket

Europe has an AI rulebook. The grid will get back to you. Download a free, practical EU AI Act checklist for small teams without handing over your email address.

An ad in my feed offered an EU AI Act checklist. Before I could find out what the law might require me to disclose, the download form wanted my contact details. A surprisingly hands-on introduction to the subject.

I’ve put together a free checklist for small teams that build or use AI. The entire procurement process is clicking a link.

The power is pending

Europe has an impressive ability to get the paperwork ready ahead of the equipment. In July 2026, the European Commission opened a call for up to seven AI Gigafactories. A month earlier, its energy roadmap pointed to delays in grid development as a major reason why big electricity users, including data centres, sit in connection queues. The announcement can go out as soon as someone approves the wording. The transformer has a different lead time.

The Dutch government restricts new data centres that meet its hyperscale criteria to two designated areas. Ireland’s energy regulator has set conditions for new connections, including a requirement to meet at least 80 per cent of annual demand with additional renewable generation in Ireland, with six years for that generation to come online. Those are national choices, and grid capacity, land and power costs are real constraints. They are also the details that determine whether the researchers invited to train a model can actually run it.

If we need a new tax, may I suggest a modest levy on every official use of the word sovereignty? We could fund a substation by lunch.

A rulebook with a rulebook

Some of the rules are easy to defend. I’d like to know when a customer service chat is run by software. I’d also like my next job application to survive something more rigorous than a model’s hunch about my CV. A disclosure for a support chatbot and safeguards for software that evaluates candidates address different problems. A good rule makes that difference clear.

Then there’s the work of finding the rule. The EU adopted a Digital Omnibus on AI to simplify implementation. The revised dates matter. Article 50’s transparency rules generally began applying in August 2026. The main rules for high-risk systems listed in Annex III, including certain recruitment tools, apply from December 2027. The comparable date for the Annex I product route is August 2028. Anyone operating from a slide deck made before July 2026 may want to check which calendar it came with.

Think of a university lab and a small company trying to turn the lab’s work into something customers can use. The Act has a specific exclusion for AI developed and put into service solely for scientific research and development. A commercial deployment raises a new set of questions. Who provides the system? What will it be used for? Whose decisions will it affect? The time spent answering those questions comes out of the same week as the work of testing the system. Large firms can assign that work to another department. At a small firm, the other department is often the same person after lunch.

A checklist you can keep

Start with one use of AI. Write down what it does, who can be affected and who is responsible. The worksheet takes you through the rules to check and gives you space to record the evidence, the open questions and who will handle them. There’s a worked example of a shop that can show customers its chatbot notice but still needs an answer from its supplier about how long it keeps chat transcripts. Both facts belong in the record.

You can download the EU AI Act checklist as a PDF without an account or email address. If you print it, you can even tick the boxes during a power cut.

No comments yet