EU AI Act & Payroll Processing:
What Swiss Employers Need to Know

Overview

The EU AI Act (Regulation EU 2024/1689) entered its critical implementation phase in 2026, with high-risk AI system requirements now in force. Although Switzerland is not an EU member state, Swiss employers are directly affected if they operate in the EU, use EU-developed HR or payroll AI tools, or process personal data of EU residents.

For Swiss businesses using automated payroll systems — particularly those certified under Swissdec ELM 5.0 — understanding the Act's scope is now a compliance priority, not a theoretical exercise.

Which Payroll AI Systems Are Classified as High-Risk?

Annex III of the EU AI Act explicitly classifies the following employment-related AI uses as high-risk:

• Automated decisions affecting recruitment, promotion, or termination
• AI-driven monitoring of employee performance or behaviour
• Systems that allocate work tasks or evaluate productivity
• Automated processing of sensitive personal data in HR decisions

Core payroll calculation (applying a rate table to a gross salary) is generally not high-risk, as it is deterministic and rule-based. However, the line blurs when AI is used for anomaly detection, retroactive payroll auditing with pattern recognition, or predictive modelling of sick leave costs.

Where Swissdec-Certified Systems Stand

Swissdec ELM-certified systems transmit salary data to Swiss social insurance authorities (AHV, BVG, SUVA) via a standardised XML interchange. The ELM protocol is deterministic: it applies fixed statutory rates and rules without learning or optimisation algorithms. As a result, the ELM data exchange layer itself does not fall under the EU AI Act's high-risk classification.

Where caution is required is in the surrounding analytics layer — dashboards that surface salary benchmarks, employee risk scores, or absence patterns using machine learning models may qualify as high-risk systems if their outputs influence employment decisions.

Documentation Requirements for Swiss Employers

If your organisation uses a high-risk AI system (or procures one from an EU vendor), you must be able to demonstrate:

• Technical documentation — how the system was trained, what data it uses, and how outputs are generated
• Human oversight mechanisms — evidence that a qualified person reviews and can override AI decisions
• Data governance records — training data provenance and bias evaluation results
• Conformity assessment — either self-assessment (for lower-risk high-risk systems) or third-party notified body review
• Incident log — records of any serious incidents or near-misses involving the AI system

Interaction with Swiss nFADP (Revised FADP)

Switzerland's own revised Federal Act on Data Protection (nFADP, in force September 2023) requires disclosure when automated individual decisions are made about data subjects. Under Art. 21 nFADP, affected employees must be able to request a human review of any automated decision. This aligns closely with — but is not identical to — the EU AI Act's human oversight mandate.

Swiss employers should align their AI governance policies to satisfy both regimes simultaneously: document the decision logic, provide an opt-out or review mechanism, and retain audit logs for at least 12 months.

What SevenSprings Recommends

Swiss Payroll Expert, SevenSprings' flagship product, uses rule-based Swiss statutory calculations (AHV, BVG, SUVA, FAK, ALV) without machine learning in the core wage engine. Our AI features — powered by Copilot Studio — are advisory only: they surface information and draft responses but do not issue decisions. Human confirmation is required for every payroll action.

We are currently preparing our EU AI Act technical dossier for the advisory components and expect to publish it in Q3 2026. Clients requiring documentation for their own compliance records can request a draft upon written request.