What the EU AI Act Really Requires From Your HR Team (and What's Coming Next)

For months, August 2, 2026 was cited everywhere as the deadline: from that date, any company using AI to screen CVs, evaluate performance, or decide on promotions would have to comply with a full package of obligations under the EU AI Act, facing fines of up to €15 million.

On June 29, the Council of the EU gave final approval to the Digital Omnibus on AI, which delays the toughest obligations for high-risk systems, where nearly all AI-powered HR tools fall, until December 2, 2027. So no, this month doesn't change much dramatically. But we wouldn't recommend shelving the topic for a year and a half either.

Here's what's worth having clear:

1. What we do have to comply with today

• The ban on using AI to infer emotions from candidates or employees through facial analysis or tone of voice, in force since February 2025.
• AI literacy: making sure anyone using these tools understands how they work and what their limitations are, also in force since February 2025.
• The transparency obligation: informing a candidate or employee when they're interacting with an AI system that influences a decision about them, in force since this August.

The question that helps you check this:

• Does everyone who goes through a selection or evaluation process at your company know AI is involved, and who tells them?

2. What's delayed until December 2027

What's pushed back to December 2027 is the heavier package: structured human oversight, verifying the provider's conformity documentation, a fundamental rights impact assessment, and logging activity for every high-risk system, meaning ATS tools with scoring, automated performance evaluation, or tools that influence promotions or dismissals.

More time isn't the same as not having to do it. It's time to build a proper process, instead of a rushed one thrown together against the deadline.

3. Why this isn't just an HR issue

So far, everything has focused on recruitment, evaluation and performance, because that's where the law puts its explicit focus and where regulatory pressure is felt first. But the ATS or evaluation tool is rarely the only place in the company where AI is already running: it tends to show up in marketing, customer service, finance and product too, often without anyone deciding that centrally.

The 2027 deadline is specific to high-risk employment systems. But the underlying question, which AI are we using, who approved it, what data does it work with, and who oversees what it does, applies just as much to the whole company, not only HR. And that question is worth answering now, deadline or not.

4. The inventory you can start now, without waiting for 2027

The first step doesn't depend on any legal deadline: knowing which tools across the company use AI, and for what, department by department, not only in recruitment or evaluation.

Worth mapping:

• Which HR, marketing, customer service, finance or product tools use AI at some point in the process?
• Does each provider have their conformity documentation ready, or does someone need to ask for it?
• Who in the company would be responsible for overseeing, and if needed, overriding a decision a system has made?
• Are any teams using AI on their own (a recruiter filtering candidates, a salesperson drafting proposals, a manager writing evaluations) without it being a formally approved tool? Responsibility doesn't disappear because the use isn't official, it still belongs to the company.
• Not using AI in any process yet? Then the question isn't about inventory, it's about the starting point: what to implement, in what order, and with what governance from day one, so nothing has to be redone once the 2027 deadlines arrive.

How we do it at DO'IN TALENT

With DO'IN AI we run exactly this exercise, but across the whole company, not just HR: where AI is already running, what governance is missing, and which priorities make sense to tackle first based on actual risk, not regulatory noise.

And if your company doesn't use AI yet, the exercise is different but just as necessary: we design the implementation from scratch, with business priorities driving it and compliance built in from the design stage, not bolted on afterwards. That avoids both the paralysis of waiting until everything's resolved and the rush of adopting scattered tools without criteria.

If you want a first snapshot before that conversation, our free AI Readiness assessment does exactly that: 15 questions about data, tools, culture and governance, built for the whole organization, not just the people department. In a few minutes you'll know where your company stands today and which priority to tackle first. Let us know and we'll set it up with you.