← HR software guides
Home → HR Software Guides → Pymetrics / Harver EU AI Act
⚠ POSSIBLE BANNED PRACTICE — Article 5(1)(f)HIGH RISK — Annex III Category 4

⚠ Urgent legal review recommended

AI systems that infer emotions in the workplace have been prohibited since 2 February 2025 under Article 5(1)(f). If Pymetrics or Harver infer emotional states or psychological traits from game-based behaviour data, continued use may constitute a prohibited practice with fines up to €35M or 7% of annual turnover. Seek specialist legal advice before the August 2026 enforcement ramp-up.

Article 5 prohibited practices apply since 2 February 2025 — not August 2026. If you are using these tools today, the prohibited practice provisions are already in force.

Pymetrics / Harver EU AI Act: Are You in the Banned Practice Zone?

Pymetrics (now part of Harver) uses neuroscience games to assess candidates' cognitive and emotional traits. This methodology sits at the intersection of high-risk AI and potentially prohibited AI under the EU AI Act. Here is what deployers need to understand and do.

Feature risk classification

FeatureClassificationLegal basis
Neuroscience games — if inferring emotional statesLIKELY PROHIBITEDArticle 5(1)(f) — emotion inference in workplace prohibited
Cognitive trait assessment — attention, memory, risk toleranceHIGH RISKAnnex III Cat.4 — AI in employment decisions
Candidate matching / rankingHIGH RISKAnnex III Cat.4
Skills-based job matching (without emotion/personality inference)HIGH RISKAnnex III Cat.4 — still requires full deployer obligations

Immediate actions for current Pymetrics / Harver users

Need help assessing your risk exposure?

The HR AI Compliance Kit includes a Pymetrics/Harver-specific Article 5 risk assessment checklist, candidate disclosure templates, and a legal review brief you can hand to counsel.

Get the HR Compliance Kit — €99 →

Article 5(1)(f) — the emotion recognition prohibition

Article 5(1)(f) EU AI Act: It is prohibited to place on the market, put into service or use AI systems that "infer emotions of natural persons in the context of the workplace and educational institutions, except where the use of the AI system is intended for medical or safety reasons."

This prohibition came into force on 2 February 2025 — eight months before the August 2026 high-risk deadline. Any AI system that infers emotional states from data (facial expressions, voice patterns, physiological responses, game behaviour patterns) in a workplace context has been prohibited for over a year at the time this guide was written.

Does Pymetrics infer emotions?

Pymetrics describes its methodology as measuring "cognitive and emotional traits" using neuroscience games. The word "emotional" in their own product description is the critical issue. If their model produces any output that characterises a candidate's emotional state, emotional regulation, or emotional reactivity — even as an intermediate step in a larger scoring process — that is potentially emotion inference under Article 5(1)(f).

Pymetrics and Harver have not published a clear public legal opinion on this question. The burden falls on you as the deployer to investigate.

What "infer emotions" means in practice

The prohibition covers any system that draws conclusions about a person's emotional state from observable data. It does not require the output to be labelled "emotion score" — if the model uses emotional responses as an input feature or produces a score that encodes emotional characteristics, that likely qualifies. The key test: does the AI system, at any point in its pipeline, make a determination about how a candidate feels or what their emotional traits are?

Harver's skills-only assessment pathway

Harver offers assessments that test specific job-relevant skills (situational judgement, workplace simulation, cognitive ability for specific tasks) without personality or emotional inference. These are still high-risk under Annex III Category 4 — but they do not face the Article 5 prohibition concern. If you need to continue using Harver, work with them to identify which assessment modules do not involve emotion or personality inference, and restrict your use to those.

Frequently asked questions

We signed a contract with Pymetrics before the EU AI Act. Are we protected?
No. EU AI Act prohibited practices apply regardless of when a contract was signed. If using the tool today constitutes a prohibited practice, you are in breach today — not from when you signed the contract. You should review whether any contractual terms allow you to exit or suspend use without penalty given the change in regulatory environment, and take legal advice on this.
Pymetrics told us they are compliant. Is that enough?
No. A vendor's claim of compliance is not a legal defence for the deployer. You need to see their legal analysis — specifically an opinion on whether Article 5(1)(f) applies to their product, and if they claim it does not apply, what the legal reasoning is. Request this in writing. If they cannot or will not provide it, that is itself a significant signal.
What are the fines if Article 5 applies?
Violations of Article 5 prohibited practices attract the highest fines in the EU AI Act: up to €35 million or 7% of worldwide annual turnover, whichever is higher. For providers, this is the most severe tier. Deployers using a prohibited AI system also face penalties — the maximum is up to €15 million or 3% of turnover for deployer violations, though deploying a prohibited system could be treated more severely.