Credit scoring AI high-risk: when does it really apply?

Anyone using a scoring system soon asks whether it counts as high-risk AI. The answer is in Annex III, point 5 of the AI Act, and it is narrower than many assume.

This article shows what points (b) and (c) cover literally, what is expressly excluded and what deployers must observe from 2 December 2027.

It covers AI systems intended to be used to evaluate the creditworthiness or credit score of natural persons. The word “intended” points to the intended purpose of the system.

Two limits are in the wording itself. Systems for detecting financial fraud are excluded. And credit checks on companies, that is legal persons, are not covered. Sole traders, freelancers and the self-employed, however, are natural persons and therefore covered.

An example from the mid-market: an online retailer has the creditworthiness of private customers assessed automatically before a purchase on invoice. The high-risk item here is the scoring system. The retailer is a deployer only if it uses the system itself. If an invoice-payment service takes on the default risk and decides itself, the service is the deployer. A pure fraud check, for example of identity or delivery address, is excluded. This is an interpretation of the wording, not an authority’s ruling. Check the intended purpose in the provider’s documentation.

If admission to purchase on invoice depends decisively on the score, Art. 22 GDPR may also apply (CJEU C-634/21, SCHUFA).

Point (c) names risk assessment and pricing in life and health insurance, each in relation to natural persons. Other types of insurance are not listed there.

That does not mean such systems are legally irrelevant. Data protection law and other rules apply regardless. High-risk status can also arise from another point of Annex III, for example if an authority uses it to decide on public benefits (point 5(a)). The profiling rule in Art. 6(3) does not change that: it applies only to systems already listed in Annex III and does not make a motor pricing model high-risk.

The table classifies typical systems by the wording of Annex III point 5. It does not replace an assessment of the specific system.

Where a system falls under Annex III, there is an exemption under Art. 6(3). It applies only if there is no significant risk to health, safety or fundamental rights and one of the four listed conditions is met. Systems that profile natural persons, however, are always considered high-risk.

Profiling is any automated processing of personal data to evaluate personal aspects, including economic situation or reliability. According to the CJEU, credit scoring of natural persons is profiling (C-634/21, para. 47). On that reading, the exemption under Art. 6(3) is therefore practically ruled out for systems under point (b).

The high-risk obligations for systems under Annex III apply from 2 December 2027 (Art. 113(3)(c) as amended by the Omnibus). Before that the date was 2 August 2026.

An overview of deployer duties is on our page about deployer obligations under the AI Act. On the question of provider versus deployer, read EU AI Act technical documentation: Who must prepare it?.

The steps are a recommendation. The regulation does not prescribe a sequence.

High-risk is not a blanket verdict for creditworthiness and insurance. What counts is who is assessed and the type of insurance.

The classification can only be evidenced if you have recorded for each system who it assesses and what it is intended for. That is also the basis for every later obligation.

This article is for general information only and does not constitute legal advice. Last updated: 19 October 2026.