French public sector HR advisers need AI literacy that goes beyond awareness: knowing what depth an agent holds, reviewing its draft administrative acts against the statute, and spotting when an employment use becomes high-risk under the EU AI Act. Careers, mobility and discipline decisions stay with them. GDPR, the CNIL and Article 4 frame the training duty.
What occupies an HR adviser in French public administration?
Careers are managed through statute. An adviser in a ministry, a regional authority or a public hospital handles grade advancement, transfers, secondments, recognition of prior service when a contract worker becomes a titular official, part-time and remote work authorisations, and the paperwork for leave on medical grounds. Almost every outcome becomes a formal administrative act that a signing authority issues.
The adviser is the person who knows which rule applies to which category of staff. Titular officials, contract staff and trainees sit under different texts, and the three branches of the public service each have their own variations. Much of the job is checking that a draft act cites the right basis and the right dates, because an act with a defective basis can be challenged and annulled.
Where do agents take on part of the work?
Agents fit the high-volume, text-heavy tasks. Each is proven at a depth on the employer's own past files.
- Seniority reconstruction (depth 1). The agent reads contracts and prior service certificates and calculates the reclassification on recruitment, showing each period and the rule applied.
- Draft administrative acts (depth 2). For routine decisions such as a part-time authorisation renewal, it drafts the act from the request and the applicable text. An adviser checks the visas and the signing authority issues it.
- Questions from staff on remote work and leave (depth 2). It drafts a reply citing the internal charter. A person sends.
- Medical committee scheduling (depth 3). Within limits the employer sets, it books slots and sends convocations, logging each one.
Sorting candidates for a mobility campaign or ranking staff for promotion is a different matter. Those touch people's careers directly and should stay with the advisers and the panels, with the agent at most organising files at depth 1.
What do advisers need to learn, and what stays with them?
The first lesson is the minimum chain. Every agent action is checked against role scope, certificate, sources, the employer's own limits and the platform guardrails, and the lowest bound decides. Advisers who know this can read a held verdict and see at once whether the issue was the source, the limit or the agent's certification. Our guide to what HR advisers should learn before working with AI agents starts here.
The second is reviewing a draft act as a lawyer would: is the legal basis right, are the dates right, is the authority competent? The agent can cite the wrong version of a decree if its sources are stale. Advisers should check the provenance stamp on each fact; a sealed reference from the official text is stronger than a note forwarded by email.
The third is correction. Our piece on how to teach HR advisers to bank corrections shows how to attach each fix to its source so the next draft improves.
What stays with the adviser: every decision on advancement and mobility, disciplinary procedures, assessments of fitness for work, and advice to staff in difficulty. These are the points where a person must answer for the outcome.
Which rules in France shape oversight and literacy?
The EU AI Act applies. Article 4 requires deployers to take measures to ensure a sufficient level of AI literacy among staff who operate AI systems. Article 26 sets deployer obligations, including assigning human oversight to people with the competence and authority to exercise it, and Article 14 describes what that oversight must achieve. The Act treats certain uses in employment, such as selecting candidates or deciding promotions, as high-risk. Public bodies are among the deployers that must carry out a fundamental rights impact assessment under Article 27 before using a high-risk system.
GDPR governs staff data. Article 22 limits decisions based solely on automated processing with significant effects, Article 35 requires a data protection impact assessment where processing is likely to be high-risk, and Article 30 requires records of processing. The CNIL supervises. Public bodies also carry their own transparency duties, including explaining to an official the basis of an individual decision.
Literacy is not one level for everyone. The guide to AI agent skills by role shows how the needs of advisers, the HR director and signing authorities differ.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
What does a 90-day training plan look like?
- Days 1 to 20: shared basics. All HR staff and the HR director cover how a governed agent decides, depth, the four verdicts and the record. Record completion per person as evidence for Article 4.
- Days 21 to 45: first use. Choose seniority reconstruction at depth 1. Three advisers compare the agent's calculation with their own on recent recruitments and note every difference with its cause.
- Days 46 to 75: reviewer training. The same advisers train to operator level, review held drafts daily, and bank corrections against the governing texts.
- Days 76 to 90: decision point. Hold a calibration on anonymised files, sign the list of approvers, and decide whether draft acts at depth 2 are ready to be proven. Update the impact assessment.
Our training plan template for HR advisers gives the fields to record each step.
Where the Academy fits
Pryme Intelligence Academy gives every HR staff member the Foundations badge, free and without expiry, which documents baseline literacy. Advisers who approve held drafts take the Operator track; its practicum asks for two weeks of real review decisions, at least 95 in every 100 correct with no missed red line. Certified Pryme Operator is valid for 12 months, so the literacy evidence stays current.
Questions people ask
Is an agent that drafts administrative acts high-risk under the AI Act?
Drafting routine acts that a person checks and signs is different from selecting or promoting staff. Whether a specific use is high-risk depends on what it decides and how; assess each use separately with counsel.
Does the Foundations badge satisfy Article 4 on its own?
Article 4 asks for a sufficient level of literacy for the context. A shared baseline helps, but advisers who approve decisions need deeper, role-specific training.
Should staff be told an agent drafted their act?
Transparency duties for public bodies point that way, and it is good practice. The adviser who signs still answers for the content.