EU AI Act high-risk systems: assess the intended use

A sector label is not a classification

Start with the intended purpose of the specific AI system. Under Article 6, the high-risk routes cover certain regulated products and uses listed in Annex III. The article also contains conditions and exceptions that must be assessed.

The fact that a system is useful, runs in a bank or assists a hospital does not settle its classification.

Compare the actual uses

• Recruitment: a system that filters candidates needs assessment against the employment category. A tool that formats a meeting agenda performs a different task.
• Credit: evaluating the creditworthiness of natural persons is a listed use, with specified exceptions. Summarising a finance policy is not the same function.
• Infrastructure: assess whether the system is intended as a safety component within the covered critical infrastructure. An advisory report about asset records is not automatically that safety component.
• Education: admissions and assessment uses can fall within Annex III. A general literature search is a different intended purpose.

These are classification prompts, not rulings on a particular deployment. Record the analysis and revisit it when the system's purpose or level of authority changes.

Separate provider and deployer work

Provider duties include the relevant conformity and quality obligations for high-risk systems. Deployer duties concern using the system appropriately, oversight and monitoring, among other requirements. A customer can acquire additional duties when it changes a system or places it under its own name.

Prepare useful evidence

A scoped review should identify the purpose, affected people, system version, evaluation results, oversight arrangements and incident process. Log relevant activity with appropriate access and retention rather than assuming every token or private model reasoning must be stored.

NeuroCluster can support agreed workflow controls and reviewable evidence. Those controls do not classify or certify the complete system.

Sources and timeline

Checked on 15 October 2026. The Commission's AI Act overview explains the revised application dates. See the central readiness page for the timeline and the official articles linked above for role and classification requirements.