AI in HR: My Guardrails After Two Tough Missions

In short: AI delivers strong value across HR — from CV screening and automated interviews to onboarding and engagement analysis — but it carries serious legal and cultural risk when used badly. The key is operational guardrails: human oversight on hiring decisions and careful handling of candidate data, applied consistently across every use case.

In 2024 I supported two HR directors on AI integration projects — one at a 350-person industrial mid-cap, the other at a 120-person tech scale-up. Both missions threw me into legal and cultural trade-offs I hadn't seen coming. Here's what I learned, along with the operational guardrails I now hand to every HR director who consults me.

Human resources is one of the functions where AI delivers the most value — and also the most risk when it's used badly. From CV screening to automated interviews, training, and engagement analysis, here's a tour of the use cases and the precautions you actually need to take.

AI-Assisted Recruiting

This is the most widespread use case — and the most sensitive:

• ATS with AI: Lever, Greenhouse, and Workday build in candidate scoring
• Profile search: LinkedIn Recruiter, SeekOut, Eightfold.ai for sourcing
• CV analysis: Textkernel, Beamery for extraction and comparison

Be careful: AI-driven CV screening absolutely has to be audited for discriminatory bias (gender, origin, age). The EU AI Act classifies certain HR uses as "high risk," with specific obligations attached. See our GDPR and AI Act checklist.

Onboarding and Training

• 360Learning, Docebo: LMS platforms with AI-personalized learning paths
• Zavvy, Leapsome: structured onboarding with automatic check-ins
• Notion AI: an onboarding knowledge base with Q&A for new hires

Engagement Analysis and Retention

• Glint (Microsoft), Culture Amp, Lattice: engagement surveys with semantic analysis of comments
• Peakon: early detection of disengagement signals

These tools handle sensitive data about how employees actually feel. Transparency with employee representatives and GDPR compliance aren't optional — they're mandatory.

Administrative HR Automation

• ChatGPT via API: drafting job descriptions, template emails, HR meeting notes
• Whisper: transcribing annual reviews (with the employee's consent)
• HRIS with built-in AI: Workday, SAP SuccessFactors, BambooHR

The Legal Framework to Respect

In France, AI use in HR processes is governed by:

• GDPR (processing the personal data of candidates and employees)
• Employment law (informing and consulting employee representatives)
• The EU AI Act (high-risk AI systems in employment)

Without a framework, these tools can expose the company to penalties and to a deterioration of the social climate. Our GDPR guide lays out the essential control points.

My Take on AI in Recruiting: Absolute Caution

At the tech scale-up, I helped evaluate an ATS with AI scoring of CVs. After the audit, we turned down the initial rollout. The reason: the vendor couldn't demonstrate that there was no gender bias in its model. Six months later, that vendor published an independent fairness audit, which let us restart the project.

The lesson: AI in recruiting isn't a technical subject, it's a legal one. The EU AI Act, adopted in 2024 and published in the Official Journal of the EU, classifies candidate-screening systems as "high risk." That brings obligations: bias audits, model transparency, the right to a human review. Without those guarantees, the legal risk lands squarely on the employer.

Onboarding: Where AI Genuinely Shines

At the industrial mid-cap, rolling out an AI knowledge base for new joiners cut the manager's average response time to onboarding questions from 4 hours to 8 minutes. New hires now find their own answers 24/7 on the standard topics (expense reports, leave, tool access), and the manager only steps in on the genuinely specific cases.

This was the AI use case that won the fastest buy-in from employees. No legal risk, measurable time savings, a better experience for new arrivals.

Engagement Analysis: With Mandatory Works-Council Consultation

At the mid-cap, deploying a semantic-analysis tool for engagement-survey comments took four months of discussion with the works council before it got the green light. The sensitive issue: the algorithms could potentially identify employees who were critical of their management, despite the anonymization.

The solution we landed on: forced aggregation to a minimum of 10 responses per cluster, a ban on extracting anything at the individual level, and a quarterly audit by the works council. Without those guarantees, the rollout would have been blocked.

The Trap of Transcribing Annual Reviews

I watched one HR director roll out Whisper to automatically transcribe annual reviews. Good intentions — a real time-saver for managers — but the project blew up at the works council, which flagged it as intrusive. Backing down publicly after you've already announced something is always costly.

My recommendation: for this kind of project, start from the social dialogue, upstream. Co-build it with the employee representatives. Pilot it on a small, voluntary scope before you generalize.

The Legal Framework in 2026: What HR Directors Need to Know

In France in 2026, AI use in HR processes is governed by: GDPR (processing of personal data), employment law (informing and consulting employee representatives), the EU AI Act (high-risk systems in employment), and the law of 24 June 2024 on partial transposition. Without a formal framework, the legal exposure is significant.

My advice: bring in an employment lawyer from the tool-evaluation phase, not after the rollout.

Our Read at Trust-Vault

AI for HR is one of the most sensitive categories we assess. The Trust Score factors in data handling (the Privacy pillar), transparency about scoring algorithms (the Transparency pillar), and security certifications (the Security pillar). See our methodology for the details.

Further reading

For a complementary implementation angle, read Prompt Engineering: The Techniques I Actually Use Daily.

For a complementary implementation angle, read AI Tools for SMEs: The Stack I Actually Deploy in 2026.