EU AI Act: Which AI Systems Are High-Risk?

What’s It About?

The European Union’s Artificial Intelligence Act establishes a binding legal framework for the use of AI systems. At its core are so-called high-risk systems, which are subject to particularly strict requirements. The compliance deadline, originally set for August 2026, has been pushed back to December 2027, giving companies extra time to prepare. Even so, the effort required to meet the rules remains substantial.

Background & Context

The EU AI Act sorts AI applications into different categories based on their risk potential. Systems that pose a high risk to safety or fundamental rights are subject to strict obligations. These include applications in sensitive areas such as biometric identification, management of critical infrastructure, human resources, and creditworthiness assessments. The regulation requires providers of such systems to register them in an EU database, produce technical documentation, and implement mechanisms for human oversight.

An important exemption is set out in Article 6(3) of the regulation: AI systems that merely perform preparatory tasks while humans make the final decision may, in certain cases, fall outside the high-risk category. However, this self-assessment option requires careful review by providers. The European Commission has published a draft guideline that provides concrete examples, though it is not legally binding.

Typical high-risk applications include AI-powered candidate screening systems, employee-monitoring algorithms, automated credit decisions, and tools used in law enforcement. Companies must complete conformity assessments by the December 2027 deadline and demonstrate that their systems meet regulatory requirements. Misjudging the risk classification can lead to legal consequences.

What Does This Mean?

Companies need to systematically inventory their AI systems and classify them by risk potential
High-risk systems require comprehensive documentation, human oversight, and regular conformity checks
The extension to December 2027 gives companies more time for implementation but does not change the scope of the obligations
The exemption for preparatory tasks requires careful consideration and legal review
Incorrect self-classification can lead to regulatory problems