AI in evaluation, promotion and dismissal: high-risk beyond hiring

Short answer: Recruitment gets the most attention, but Annex III of the AI Act goes further. AI that helps decide on working conditions, promotion, termination, task allocation and performance evaluation is high-risk too. That brings performance dashboards, productivity scores and workforce-management systems many employers already run under the same heavy regime — with the GDPR on top.

What falls under Annex III

The regulation explicitly names AI systems used to make or influence decisions on:

• working conditions, promotion and termination of an employment relationship;
• the allocation of tasks based on behaviour or personal traits;
• monitoring and evaluating the performance and behaviour of workers.

It does not matter whether the system decides itself or "merely" makes a recommendation a manager adopts: decision support is caught too.

The employer's duties

As deployer (Art. 26) the same core duties apply as in recruitment:

• Human oversight (Art. 14): a human assesses and can set aside the outcome — no algorithm may demote or dismiss anyone unchecked.
• Inform the affected workers (and where applicable their representatives) that a high-risk system is used.
• Relevant, representative input data and monitoring for skewed outcomes.
• Certain employers must first carry out a fundamental-rights assessment (FRIA).

The GDPR: Article 22 for heavy decisions

Dismissal, a refused promotion or a sanction are decisions with significant effects. Where such a decision comes solely from an automated process, Article 22 GDPR applies: a valid basis, transparency and a right to human intervention. In practice this means an AI judgement may never be the last word.

The trap: monitoring that quietly decides

Many monitoring tools not only collect data but turn it into scores that drive promotions and reviews. At that point "measuring" shifts to "deciding" — and thus into high-risk. The line is not the technology but the use.

What to do

• Inventory your performance, planning and workforce systems, not just your recruitment tools.
• Determine per system whether it makes or influences decisions about people.
• Secure a human final judgement for every consequence to an individual.
• Document the logic and keep logs — you need them in a complaint or supervisory request.

The AI Act protects the whole employment relationship, not just the front door. Getting only recruitment right covers half the risk.