How to Classify Credit and Insurance AI Under the EU AI Act

Current as of October 29, 2026. This guide draws on the EU AI Act (Regulation (EU) 2024/1689) as amended by Regulation (EU) 2026/1744, and on two sets of European Commission guidelines, one of them still a draft, as published on that date. How it applies depends on the firm and the system; it is not legal advice.

Two Annex III entries in the EU AI Act reach directly into banking and insurance: AI that judges individuals' creditworthiness, and AI that assesses risk and sets prices for individuals in life and health insurance. Similar-looking systems may fall outside: a fraud model under the Act's own carve-out and, in the Commission's draft guidelines, a pricing engine fed by a separate score, a claims tool or a model that rates companies.

Below is a decision sequence for each system; unless a step quotes the Act or a Commission text, it is our recommendation, labeled (interpretation).

On scope: the AI Act is a regulation, binding and directly applicable in every Member State. The Commission's guidelines on the definition of an AI system and its draft classification guidelines each say "The Guidelines are not binding"; authoritative interpretation rests with the Court of Justice of the European Union.

Step 1: Confirm the tool is an AI system under Article 3(1)

Article 3(1) defines an AI system as a machine-based system, designed to operate with varying levels of autonomy, that "infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments".

The Commission's definition guidelines, which are final but not binding, place outside it basic data processing that operates on fixed human-programmed rules, and systems used to improve mathematical optimization or to accelerate and approximate traditional, well established optimization methods, such as linear or logistic regression methods, noting that use in a consolidated manner for many years can indicate a system does not go beyond basic data processing. That is not a blanket exemption for scorecards: a machine learning credit model needs its own assessment, and the reasoning should be recorded either way (interpretation).

Step 2: Test intended purpose against Annex III point 5(b) for individual credit

The test turns on intended purpose, which Article 3(12) defines as the provider's intended use, including the specific context and conditions of use.

Four distinctions, drawn from the draft guidelines (interpretation):

• Natural persons. The draft counts sole traders and the self-employed as natural persons, but places outside 5(b) company-rating models, the owner backing a company loan (the company being the primary beneficiary), and a small-business owner's score built only from business data.
• Scoring versus pricing. The draft says a pricing system "is not covered by the use case listed in point 5(b) of Annex III, even if it relies on data obtained from a creditworthiness assessment or credit-scoring." A system combining both in one integrated process does qualify.
• Fraud detection. The draft reads the Act's fraud carve-out narrowly: "For the exception to apply, fraud detection must be the main intended use of the AI system, preceding all other purposes for which the AI system may be used."
• Prudential models. Recital 58 excludes AI provided for by Union law for prudential purposes to calculate banks' and insurers' capital requirements. The draft limits that to systems used solely for capital: a model that also scores individuals or assesses their creditworthiness stays high-risk even if it also produces IRB internal ratings.

The draft also limits 5(b) to essential private services, listing margin trading and leisure or travel loans as non-essential. That limit comes from the heading of point 5 and Recital 58 as the draft reads them, not from the words of 5(b), so treat it as a position to document rather than a safe harbor (interpretation). The draft adds that it is for the provider to show that its system is solely intended for non-essential services, and that a system intended for both essential and non-essential services will be classified as high-risk.

Step 3: Test life and health insurance underwriting and pricing under point 5(c)

Point 5(c) covers life and health insurance; the draft says "Other insurance services (e.g. car or motor insurances, home insurance etc.) are not covered by the use case." A system need not do both risk assessment and pricing. In the draft's reading, credit life, private long-term care, accident and sickness cover, and personal pension products in so far as they can significantly affect a person's livelihood in old age count; insurance-based investment products do not.

Unlike 5(b), it has no fraud exception. A fraud feature inside an underwriting system does not take it out. The draft places claims management outside 5(c), and product design where no individual's risk profile is evaluated. Map each insurance use to one of those before classifying it (interpretation).

Step 4: Check the four Article 6(3) derogation conditions, narrowly

Article 6(3) treats an Annex III system as not high-risk where it does not pose "a significant risk of harm to the health, safety or fundamental rights of natural persons, including by not materially influencing the outcome of decision making." At least one condition must be met: a narrow procedural task; improving a completed human activity; detecting deviations from prior decision-making patterns without replacing or influencing a completed human assessment absent proper human review; or a preparatory task.

The draft guidelines add limits: "As Article 6(3) represents an exception from rules aimed at (among others) protecting fundamental rights, the conditions must be interpreted narrowly." Human review alone does not change the answer: "Since human involvement cannot change the purpose and area in which a system is intended to be used, it has no effect on the classification of the system as high-risk under Article 6(2) AI Act." And "To avoid circumvention of the high-risk classification rules by system design, split architectures are assessed as a whole."

Step 5: Apply the profiling rule before relying on any derogation

Article 6(3)'s profiling sentence overrides all four conditions. Profiling has its GDPR meaning: in the draft's breakdown, automated processing of personal data to evaluate an individual's personal aspects.

A system that scores an applicant's ability or willingness to repay, or an insured person's risk, evaluates personal aspects by design, so the derogation will rarely be open for the core 5(b) and 5(c) systems (interpretation). It matters more for support tools, such as one that reads payslips into fields without evaluating the applicant (interpretation).

Step 6: Decide whether the firm is the provider or the deployer, including under Article 25

The Act defines a provider as one that develops an AI system, or has one developed, and "places it on the market or puts the AI system into service under its own name or trademark", and a deployer as one "using an AI system under its authority", other than in personal non-professional use. A bank that builds and uses its own scoring model is likely its provider; one using a vendor's model unchanged is a deployer (interpretation).

Under Article 25(1), a deployer becomes the provider of a high-risk system if it puts its name or trademark on it, substantially modifies it, or changes a system's intended purpose so that it becomes high-risk. Adapting a vendor's general-purpose model to score individual applicants is the case to watch (interpretation).

Step 7: Record the outcome and plan for December 2, 2027

If the system is not high-risk under Article 6(3), Article 6(4) applies: "A provider who considers that an AI system referred to in Annex III is not high-risk shall document its assessment before that system is placed on the market or put into service." The same paragraph adds: "Such provider shall be subject to the registration obligation set out in Article 49(2)." When that registration applies is covered in our guide to building an AI inventory. The draft expects the assessment to state the intended purpose, the condition relied on and why the system does not profile. "Upon request of national competent authorities, the provider shall provide the documentation of the assessment."

If it is high-risk, the provider carries the requirements and obligations in Sections 2 and 3 of Chapter III from December 2, 2027, the date Regulation (EU) 2026/1744 set for Sections 1 to 3. Under Article 27(1), deployers of 5(b) and 5(c) systems must assess the impact on fundamental rights before first use. Under Article 26(5), a financial institution's deployer monitoring duty is deemed met by complying with internal governance rules under Union financial services law. That covers only the monitoring duty. The other deployer obligations in Article 26 remain, including using the system in line with its instructions for use, assigning human oversight to competent staff, keeping the logs, and, under Article 26(11), informing the natural persons "that they are subject to the use of the high-risk AI system"; Article 27(3) also requires the deployer to notify the market surveillance authority of the assessment's results.

As amended, Article 111(2) applies the Act to high-risk systems placed on the market or put into service before the Chapter III application date "only if, as from that date, those systems are subject to significant changes in their designs." For credit scoring, the Commission's draft guidelines say that "the definition of material change to the credit-scoring within the IRB Approach as set out in the relevant Union legislation will be relevant" to whether a 5(b) system has had a significant change in design. Beyond that pointer, how routine recalibration or retraining counts is not settled in the texts reviewed; decide it with counsel and record the rationale before December 2, 2027 (interpretation).

Checklist

Items restate the recommendations above (interpretation).

• Each system has a reasoned answer on whether it is an AI system
• Credit systems are tested for natural persons, pricing, fraud, prudential-only use and essential services
• Insurance uses are mapped to underwriting, pricing, claims or product design
• Any derogation names its condition, assesses the whole system and records why it does not profile
• The firm's role is recorded, with Article 25 triggers reviewed at each vendor change
• High-risk systems have a plan for December 2, 2027, and legacy systems a rule for significant design changes
• Classifications are revisited when final Article 6(5) guidelines are adopted

How we sourced this guide: every factual statement in it is drawn from published rules, regulatory guidance and other public documents, and we keep a full set of sources and citations for it. If you would like them, or want to question a point, get in touch through our contact page and we will share them.