AI in onboarding under the EU AI Act: the transition from Article 27 candidate notice to 4(b) worker management

Onboarding is for HR teams traditionally "the quiet part": paperwork, introductions, first days. With the rollout of AI chatbots for new employees, AI matching for buddy systems, personalized learning paths and integration recommendations, that story has changed. For the EU AI Act onboarding also represents a specific transition: the candidate with Article 27 information duty becomes a worker with Annex III point 4(b) obligations.

This post explains where AI sits in modern onboarding, why the transition is legally a separate moment, and what HR teams need to arrange before the first workday.

Where AI is deployed in onboarding

The modern onboarding stack increasingly covers:

• AI chatbots for new employees — Microsoft Viva onboarding companion, custom GPT bots for HR questions, 24/7 self-service
• Buddy/peer matching AI — algorithms pairing new employees with experienced colleagues based on role, personality, location or expertise
• Personalized learning paths — LinkedIn Learning, Cornerstone, Degreed with AI recommendations for relevant training based on role and skill gaps
• Document AI for paperwork — automated processing of ID documents, certificates, contracts
• Goal-setting and 30-60-90 plan AI — AI suggests SMART goals for the first 90 days based on role and team
• Integration recommendations — AI suggests relevant internal projects, communities or stakeholders for the new employee

Not everything in this list is necessarily high-risk. But the transition from candidate to worker brings new obligations HR teams often underestimate.

The legal transition: Article 27 → Annex III 4(b)

For candidates under the AI Act Article 27 information duty applies: they must know AI scores or assesses them during the selection procedure. Once the contract is signed and the employee starts, the legal position shifts:

• From candidate to worker — different rights under GDPR, labor law, and (in NL) works council representation
• From 4(a) to 4(b) — if AI now feeds decisions on work opportunities, development or assessment, it falls within worker management high-risk
• From Article 27 candidate notice to worker information — worker information duty is broader and periodic, not one-off
• New worker representation involvement — for 4(b) systems works council often has approval rights

In practice many onboarding AI tools are immediately after hire already in 4(b) territory. An AI chatbot that answers new employee questions is logistical. An AI that does buddy matching or offers personalized learning based on inferred skills can already fall within 4(b) — depending on whether the output feeds management actions or worker assessments.

When is onboarding AI within 4(b)

The practical reading:

• Chatbots for Q&A (HR policy, leave, payroll) — logistical. Outside 4(b).
• Buddy matching AI — depends on impact. A recommendation the worker can accept/decline themselves: often outside 4(b). A match that affects manager allocation: 4(b).
• Personalized learning paths based on inferred skills — if learning enrollment is worker-driven only: lighter. If it feeds obligations or assessment input: 4(b).
• 30-60-90 goal-setting AI — if goals go directly into assessment system: 4(b).
• Document AI on contracts and ID — administration. Usually outside 4(b).
• Integration and project recommendations — based on worker visibility. Person-oriented but usually not decisive for work opportunities.

Step-by-step for onboarding AI dossier

What to do now

For HR teams with onboarding AI: design the transition from Article 27 to 4(b) explicitly. Build an onboarding AI inventory, separate logistical from decision-influencing, arrange worker information duty for the first workday, and document via the HR AI hub and the HR AI Evidence Pack. For the worker phase following onboarding: see the separate posts on performance reviews and worker monitoring.