The AI Act and recruitment tests: what changes for HR leaders in 2026

The European regulation on artificial intelligence (AI Act) came into force in August 2024, and its obligations apply in successive waves until 2027. For HR departments, one point deserves immediate attention: AI systems used to recruit, assess or promote people are classified as high-risk systems. Concretely, what does this change for your candidate assessment tools?

Why recruitment is classified as "high risk"

The AI Act (EU Regulation 2024/1689) sorts AI systems into risk levels: prohibited practices, high risk, limited risk, minimal risk. Annex III of the regulation explicitly places in the high-risk category the systems used for recruiting and selecting people, notably to target job ads, screen applications and evaluate candidates.

The legislator's logic is simple: a hiring decision affects a person's access to employment, and therefore to their livelihood. An automated system that influences this decision must meet reinforced requirements of transparency, robustness and human control.

One point is even stricter: emotion recognition in the workplace is among the practices prohibited by Article 5 of the regulation, except for narrow safety or medical exceptions. A tool that claims to infer a candidate's emotional state from their voice or face during a video interview enters a red zone, not merely a grey one.

What the regulation requires from high-risk systems

For high-risk systems, the AI Act imposes obligations that fall first on the tool's provider, but also on the company deploying it. The main families of requirements:

• Risk management: the provider must identify, assess and mitigate the system's risks throughout its life cycle.
• Data governance: training and testing datasets must be relevant, representative and managed to limit bias.
• Documentation and transparency: the deploying company must have documentation explaining what the system measures, how, and with what limitations.
• Human oversight: the system must be designed so that a human can understand its outputs, challenge them, and so that no one is subject to a fully automated decision.
• Accuracy and robustness: the system must reach an appropriate level of performance and maintain it over time.

On the employer side, two reflexes apply right now: informing candidates and employee representatives when a high-risk AI system is used in the process, and ensuring that a trained person genuinely supervises the results instead of validating them mechanically.

The uncomfortable question: is your assessment tool ready?

Many tools on the market were designed before this regulatory framework existed. For an HR director, compliance is not verified in the sales brochure but in the vendor's precise answers. Five questions to ask:

1. What scientific framework does the system rest on, and is that framework publicly available? A published, documented framework can be verified. A "proprietary AI" without a published framework cannot.
2. What exactly does the system measure, and what does it not measure? The documentation must distinguish what is assessed (behaviours in situation, for example) from what is inferred.
3. Does the system analyse voice, face or emotional signals? If so, the Article 5 question arises directly.
4. How is human oversight organised? Is the result presented as a decision aid, with its limitations, or as a verdict?
5. What data was used to calibrate the system, and how is bias monitored?

A solid vendor answers these five questions in writing. A fragile vendor answers with a product demo.

The Rising Up approach: compliance by design

Rising Up did not have to reinvent itself for the AI Act, because its initial design choices point in the same direction as the regulation.

A public framework. The 18 core skills framework is published and available: families, definitions, observable behaviours. What the Soft Skills Scan measures is documented in black and white, which directly addresses the transparency requirement. The assessment mechanics remain proprietary, as the regulation allows, but the scientific framework itself is verifiable.

A decision aid, never an automated decision. The Soft Skills Scan produces a score and a behavioural analysis that the manager or recruiter interprets. The tool informs the human decision, it does not replace it. This is exactly the human-oversight logic that the AI Act imposes on high-risk systems.

No facial or vocal emotion analysis. The assessment relies on written, contextualised professional scenarios, not on inferring emotional states from biometric signals.

An audited scientific framework. Recognised as Deeptech by the European Innovation Council (EIC) in 2023, holder of the Responsible Digital label (NR1) and of the French CIR research accreditation, Rising Up grounds its method in a documented research programme, led by its scientific team in collaboration with researchers from the CNRS, ENS-PSL and Université Paris Cité.

For institutions and companies that want to go further, the certification of behavioural competencies rests on this same documented framework: what is certified is defined, measurable and traceable.

What to remember for 2026

The AI Act timeline is gradual, but the direction is set: candidate assessment tools will have to demonstrate what they measure, how, and under what human control. Companies that recruit therefore have every interest in auditing their tools now, rather than discovering in 2027 that one link in their process does not hold.

This regulatory movement joins a deeper market movement: according to APEC, 41% of companies recruiting executives now formally assess soft skills, up 9 points in one year. The more behavioural assessment spreads, the more the scientific and legal quality of the tools becomes a selection criterion.

The question is no longer "should we assess soft skills?" but "with which tool can you justify it, to a candidate as well as to a regulator?". A recruitment test grounded in cognitive science, backed by a public framework and designed as a decision aid, answers both requirements at once.

To go further on the role of AI in HR processes, read our analysis on AI and recruiters.