The EACB welcomes the opportunity to contribute to the European Commission’s consultation on the draft Guidelines for the classification of high-risk AI systems under Art. 6 of the AI Act. The EACB supports the objective of providing greater legal certainty and practical guidance and welcomes the Commission’s efforts to accompany the Guidelines with practical examples.
At the same time, the EACB stresses that the classification of high-risk AI systems should remain firmly anchored in the AI Act’s risk-based and proportionate approach and that the classification should depend on the actual function and intended purpose of the system, rather than on its mere technical connection to broader business processes. The Guidelines should provide clearer criteria for distinguishing AI systems that genuinely influence decisions affecting individuals from systems performing preparatory, operational, governance or compliance-related functions.
Particular attention is given to the financial services use case under Annex III(5)(b). The EACB calls on the Commission to clarify that high-risk classification should apply only where an AI system is genuinely intended to assess the creditworthiness of a natural person or establish a credit score as a condition for accessing an essential private service. AI systems used for ancillary or support functions should not fall within scope merely because they are technically or operationally connected to a credit-related workflow.
Finally, the EACB also calls for greater legal certainty regarding the definition of an AI system and the treatment of established statistical methodologies. In particular, the EACB urges the Commission to confirm that traditional deterministic statistical techniques, including linear and logistic regression models based on predefined variables and fixed coefficients, fall outside the scope of the AI Act. These models are transparent, explainable and already subject to extensive governance, validation and supervisory requirements in the financial sector. Clarifying their treatment would reinforce the risk-based approach of the AI Act and avoid capturing longstanding statistical tools that do not exhibit adaptive behaviour or autonomous inference.