F3 / Human resources

Equipping HR without letting AI decide careers.

HR teams are swamped by applications, repetitive questions and scattered skills data, and AI is sold to them as the answer. It can help, provided you define what a ‘skill’ or a ‘qualified candidate’ is, keep the decision human and follow rules stricter than in any other function.

Submit a brief

Three concrete uses

01Shortlisting applications without automatically rejecting anyone
AI Act level · High risk
F3 · Shortlisting applications without automatically rejecting anyone
1. 01Written criteriaJob-related, approved before posting
2. 02Candidates informedBefore any processing
3. 03Reasoned summaryCV elements, criteria cited
4. CheckDecision by the recruiter
5. 04Recruiter’s decisionEvery application reviewed
The problem
A job ad attracts hundreds of applications and the recruiter has no time to read them all with equal care.
What the AI does
AI extracts from each CV the elements linked to the ad’s written criteria and presents a reasoned summary. It rejects no candidate, does not score personality and analyses neither voice nor face.
Data needed
Job ad, approved criteria grid, CVs and cover letters from candidates informed beforehand.
Who signs off
The recruiter reviews every application, including those AI did not highlight, and alone decides what happens next.
High-risk (EU · AI Act, Annex III, point 4), with obligations applying from 2 December 2027; informing candidates beforehand is already required (France · Labour Code, Article L1221-8).
02Answering employees’ questions about their entitlements
AI Act level · Transparency obligation
F3 · Answering employees’ questions about their entitlements
1. 01Approved corpusDated agreements and notes
2. 02Employee questionAssistant labelled as AI
3. 03Sourced answerPassage cited, dated version
4. CheckHandover to an HR officer
5. 04Individual caseTaken over by an officer
The problem
HR answers the same questions every day about leave, health cover or company agreements, and the answers vary depending on who replies.
What the AI does
An assistant answers from company agreements, the collective agreement and internal notes, citing the passage used. It settles no individual case and hands over to an HR officer as soon as a decision is at stake.
Data needed
Company agreements, the collective agreement, internal rules and up-to-date HR notes, with their validity dates.
Who signs off
The HR manager approves the corpus and standard answers; an HR officer takes over every individual case.
Transparency (EU · AI Act, Article 50): employees must know they are talking to an AI. The use would become high-risk if it served to evaluate employees or allocate their tasks.
03Mapping skills to build the training plan
AI Act level · Minimal risk
F3 · Mapping skills to build the training plan
1. 01Common frameworkOne skill, one definition
2. 02Linked sourcesJob descriptions, reviews, catalogue
3. 03Aggregated viewGaps by job family, no names
4. CheckApproval by the training manager
5. 04Training planDecided by HR
The problem
Skills are described in job descriptions, reviews and the training catalogue, in different words. The training plan is built without an overall view.
What the AI does
AI maps these sources to a common framework and produces an aggregated view of gaps by job family or team. It scores no employee and proposes no promotion, assignment or sanction.
Data needed
Skills framework, job descriptions, training catalogue and aggregated review data.
Who signs off
The training manager approves the map before any budget decision.
Minimal as long as the analysis stays aggregated. Used to evaluate, promote or assign employees, it would be high-risk (EU · AI Act, Annex III, point 4).

To define before launching AI

This is meaning debt: if these words lack a single definition and an owner, no agent will give a reliable answer, and no control will be possible.

“Skill”Certified know-how, self-declared by the employee or inferred from a job title: AI will treat these three sources as equivalent unless someone tells them apart.
“Qualified candidate”Only job-related criteria, written before the ad is posted, can ground a shortlist; otherwise the model will reproduce past choices.
“Length of service”In the company, in the group or in the role: the answer can change certain entitlements, and therefore the assistant’s answers.
“Performance”Objectives met, the manager’s appraisal or activity metrics: without a written definition, an analytics tool drifts into individual evaluation, which is high-risk.

What applies to you

EU · AI Act (Annex III)Recruitment, selection, promotion or termination decisions, task allocation and employee evaluation are high-risk, with obligations applying from 2 December 2027 (Regulation (EU) 2026/1744). Employers will then have to inform workers’ representatives and affected workers before putting a system into service (Article 26).
EU · AI Act (Article 5)Since 2 February 2025, using an AI system to infer a person’s emotions in the workplace has been prohibited, except for medical or safety reasons. The European Commission’s (EU) guidelines include recruitment within this scope.
France · Labour CodeIn companies with at least 50 employees, the works council (CSE) is consulted before new technologies are introduced (Article L2312-8) and informed before recruitment aid methods are used (Article L2312-38). Candidates and employees are informed in advance of the methods used on them (Articles L1221-8 and L1222-3).
EU · GDPRA decision with legal or similarly significant effects cannot, save for exceptions, be based solely on automated processing (Article 22). Profiling for HR management is on the list of processing operations requiring an impact assessment (France · CNIL; Article 35).

What to measure

Share of applications reviewed by a recruiter, including those AI did not highlight
Gap in shortlisting rates between women and men, measured at each campaign
Share of HR assistant answers taken over or corrected by an officer

What Clairancy does

A · Framing and decision
Classify each HR use under the AI Act and the Labour Code, write criteria and definitions, and plan works council information and consultation.
C · Sovereign AI and data
Design an HR assistant sourced from your agreements, hosted under your control, and test its answers on real questions before opening it to staff.
D · Governance and compliance
Prepare the impact assessment, the information notices for candidates and employees, the works council file and the logs that prove human oversight.

Concepts to understand

Read on the blog

Frequently asked questions

Other functions

Can we analyse candidates’ emotions in video interviews?

No. Inferring emotions in the workplace has been prohibited since 2 February 2025 (EU · AI Act, Article 5), and the European Commission includes recruitment in this ban. Only medical or safety reasons are exempt.

Must the works council be consulted for a mere pilot?

Often, yes. A French court suspended AI tools that had been tested for months without consultation, finding that the pilot went beyond mere experimentation (France · Nanterre Judicial Court, 14 February 2025). Inform and consult before opening the tool to staff.

Does the AI Act delay suspend our recruitment obligations?

It only moves high-risk obligations to 2 December 2027. Informing candidates beforehand, the GDPR and works council consultation already apply.

Independent analysis, not legal advice. Up to date as of 29 October 2026.

Equipping HR without letting AI decide careers.

Submit a brief

Three concrete uses

01Shortlisting applications without automatically rejecting anyone

AI Act level · High risk

F3 · Shortlisting applications without automatically rejecting anyone

01Written criteriaJob-related, approved before posting

02Candidates informedBefore any processing

03Reasoned summaryCV elements, criteria cited

CheckDecision by the recruiter

04Recruiter’s decisionEvery application reviewed

The problem: A job ad attracts hundreds of applications and the recruiter has no time to read them all with equal care.
What the AI does: AI extracts from each CV the elements linked to the ad’s written criteria and presents a reasoned summary. It rejects no candidate, does not score personality and analyses neither voice nor face.
Data needed: Job ad, approved criteria grid, CVs and cover letters from candidates informed beforehand.
Who signs off: The recruiter reviews every application, including those AI did not highlight, and alone decides what happens next.

High-risk (EU · AI Act, Annex III, point 4), with obligations applying from 2 December 2027; informing candidates beforehand is already required (France · Labour Code, Article L1221-8).

02Answering employees’ questions about their entitlements

AI Act level · Transparency obligation

F3 · Answering employees’ questions about their entitlements

01Approved corpusDated agreements and notes

02Employee questionAssistant labelled as AI

03Sourced answerPassage cited, dated version

CheckHandover to an HR officer

04Individual caseTaken over by an officer

The problem: HR answers the same questions every day about leave, health cover or company agreements, and the answers vary depending on who replies.
What the AI does: An assistant answers from company agreements, the collective agreement and internal notes, citing the passage used. It settles no individual case and hands over to an HR officer as soon as a decision is at stake.
Data needed: Company agreements, the collective agreement, internal rules and up-to-date HR notes, with their validity dates.
Who signs off: The HR manager approves the corpus and standard answers; an HR officer takes over every individual case.

Transparency (EU · AI Act, Article 50): employees must know they are talking to an AI. The use would become high-risk if it served to evaluate employees or allocate their tasks.

03Mapping skills to build the training plan

AI Act level · Minimal risk

F3 · Mapping skills to build the training plan

01Common frameworkOne skill, one definition

02Linked sourcesJob descriptions, reviews, catalogue

03Aggregated viewGaps by job family, no names

CheckApproval by the training manager

04Training planDecided by HR

The problem: Skills are described in job descriptions, reviews and the training catalogue, in different words. The training plan is built without an overall view.
What the AI does: AI maps these sources to a common framework and produces an aggregated view of gaps by job family or team. It scores no employee and proposes no promotion, assignment or sanction.
Data needed: Skills framework, job descriptions, training catalogue and aggregated review data.
Who signs off: The training manager approves the map before any budget decision.

Minimal as long as the analysis stays aggregated. Used to evaluate, promote or assign employees, it would be high-risk (EU · AI Act, Annex III, point 4).

What applies to you

EU · AI Act (Annex III)Recruitment, selection, promotion or termination decisions, task allocation and employee evaluation are high-risk, with obligations applying from 2 December 2027 (Regulation (EU) 2026/1744). Employers will then have to inform workers’ representatives and affected workers before putting a system into service (Article 26).

EU · AI Act (Article 5)Since 2 February 2025, using an AI system to infer a person’s emotions in the workplace has been prohibited, except for medical or safety reasons. The European Commission’s (EU) guidelines include recruitment within this scope.

France · Labour CodeIn companies with at least 50 employees, the works council (CSE) is consulted before new technologies are introduced (Article L2312-8) and informed before recruitment aid methods are used (Article L2312-38). Candidates and employees are informed in advance of the methods used on them (Articles L1221-8 and L1222-3).

EU · GDPRA decision with legal or similarly significant effects cannot, save for exceptions, be based solely on automated processing (Article 22). Profiling for HR management is on the list of processing operations requiring an impact assessment (France · CNIL; Article 35).

What Clairancy does

A · Framing and decision

Classify each HR use under the AI Act and the Labour Code, write criteria and definitions, and plan works council information and consultation.

C · Sovereign AI and data

Design an HR assistant sourced from your agreements, hosted under your control, and test its answers on real questions before opening it to staff.

D · Governance and compliance

Prepare the impact assessment, the information notices for candidates and employees, the works council file and the logs that prove human oversight.