AI is already in your payroll software. Does your team know what the AI Act expects?

Article 4 of the EU AI Act has applied since 2 February 2025 and requires organisations that use AI systems to ensure a sufficient level of AI literacy among the staff who work with those systems. For payroll and HR service providers this is not an abstract rule: AI features increasingly appear in payroll software, from anomaly detection in the pay run to assistants that answer regulatory questions. There is no separate fine attached to Article 4. The Digital Omnibus has been formally adopted but not yet published; the adopted text changes Article 4 into a duty to take measures that support the development of AI literacy. The current text remains in force until the amendment enters into force. The expectation that your staff can demonstrably handle AI therefore stands, especially in a profession built on wage and personal data.

In practice, several roles in a payroll organisation already use AI. Payroll administrators work with AI checks that flag anomalies in the pay run, consultants use chatbots to prepare answers to collective agreement and legal questions, and service desks summarise tickets with generative AI. What matters is the boundary with high-risk use: AI deployed for recruitment, selection, promotion or termination decisions falls under Annex III of the AI Act, for which the adopted amendment moves most standalone obligations to 2 December 2027 once it enters into force. Payroll administration itself usually sits outside that scope, but payroll and HR service providers often operate exactly on that boundary. Knowing where it lies tells you when extra requirements apply and when human review of AI output is enough.

Provable means more than an attendance sheet. The organisational evidence records show who completed what, when, time spent, and assessment results; the certificate confirms completion. For a payroll organisation this is doubly relevant, because clients, accountants and regulators increasingly ask how carefully AI and personal data are handled. A review-ready Article 4 dossier bundles this evidence, so that when a client or auditor asks, you do not have to reconstruct what happened but can simply show the report.

For training providers serving payroll professionals there are three delivery models. Co-brand: the provider offers the training under both names and keeps the client relationship. For white-label delivery, brand and delivery scope are agreed in advance; removing every LearnWize reference, including certificate and registry branding, is not a standard promise. SCORM: the module is delivered as a SCORM 1.2 package and runs in the provider's or the client's own LMS. The SCORM variant has been tested end-to-end in Moodle; behaviour in other learning environments depends on their SCORM 1.2 implementation. In every model the provider brings the network, the academy and any recognition from a professional body, while LearnWize supplies content based on official EU sources plus organisation-wide reporting with learner records.