CV screening gets attention because applicants are visible; performance reviews slip under the radar. But AI in performance management hits Annex III point 4(b) of the EU AI Act directly — and the impact on workers is in many respects greater than on candidates. Assessment decisions determine promotion, pay, development and ultimately continuity. If AI helps drive those decisions, you are in deployment territory that requires a separate route.
This post explains where AI sits in modern performance tools, why point 4(b) works differently than 4(a), and what you can arrange today before your workers notice.
Where AI sits in performance reviews today
Modern performance platforms (including parts of SAP SuccessFactors, HiBob, and increasingly BambooHR) contain AI at multiple levels:
- Feedback generation — AI suggests text for managers writing feedback, based on project data and historical reviews
- Calibration suggestions — AI helps teams score consistently across managers, with pattern detection and bias correction
- Performance prediction — AI predicts future performance based on history, project results and peer feedback
- Goal-setting support — AI suggests SMART goals based on role and team context
- Skill gap analyses — comparison of employee skills with role requirements, with development recommendations
- Succession planning — AI identifies candidates for key positions based on performance and potential signals
At first glance supporting work. But this AI output affects manager decisions about pay, promotion, development and ultimately continuity. That is exactly what Annex III point 4(b) covers.
Why point 4(b) works differently than 4(a)
Point 4(a) (recruitment) and point 4(b) (worker management) are both high-risk under the AI Act, but the practical context differs:
| Element | 4(a) Recruitment | 4(b) Worker management |
|---|---|---|
| Who is affected | Candidates | Existing workers |
| Information duty | Article 27 candidate notice | Article 27 + worker representation |
| FRIA context | Often recruitment-wide | More specific per use case |
| Bias risk | Statistical | More personal, years of impact |
| Worker rights | Privacy rights, no contract | GDPR + employment contract |
The heaviest practical differences sit in two things: worker representation (in NL the OR with WOR instemmingsrecht) has approval rights for systems that monitor or assess workers. And cumulative impact: a candidate can be rejected and move on, a worker lives for years with the assessments AI helps shape.
When is performance AI within point 4(b)
Not every AI function in an HR platform automatically falls under 4(b). The practical reading:
- AI that generates feedback text the manager can adjust — context-dependent. If the manager is genuinely free and the text only serves as a starting point, it likely stays outside 4(b). If the generated feedback goes directly into dossier or is used as calibration input, it tips.
- AI that gives calibration suggestions (bias correction between managers) — directly affecting individual worker scores. Within 4(b).
- AI that predicts performance or calculates risk scores — determining for promotion/development decisions. Within 4(b).
- AI that identifies skill gaps for development — if purely informational and the worker uses it themselves: lighter. If it drives manager decisions on development budget: 4(b).
- AI that nominates succession candidates — direct influence on work opportunities. Within 4(b).
Like with CV screening: it is layered. Not every performance AI is high-risk, but much more than organizations realize is.
Step-by-step for your performance AI dossier
Start with a module audit of your HR stack
A lot of performance AI sits hidden in broader HRIS systems (Talent Intelligence Hub, succession modules, calibration tools). Map the tool landscape before classifying.
Treat worker representation as a parallel track
For Dutch employers with works councils: don't wait until your AI Act dossier is complete. WOR approval for monitoring/assessment systems is a separate legal basis.
Document in HR AI Evidence Pack under 4(b) section
Use the HR AI Evidence Pack — the template has a separate 4(b) module for worker management context.
Test your AI against the Article 6(3) filter
Interactive self-assessment, updated for the Commission guidelines of 19 May 2026. 9 steps, personal report with reasoning, vendor questions and next steps.
Frequently asked questions about performance AI and the AI Act
Practical questions for HR directors on Annex III point 4(b) deployments.
What to do now
For HR directors wanting to understand AI in performance reviews for the AI Act: start with a module audit this month, schedule a worker representation conversation in parallel, and use the HR AI Evidence Pack for your 4(b) dossier. For the broader context of AI in work and worker management: the HR AI hub.