Annex III point 5 covers AI systems intended to be used by or on behalf of public authorities to evaluate eligibility for essential public assistance benefits and services, including healthcare, and to grant, reduce, revoke or reclaim such benefits; to evaluate creditworthiness or establish a credit score, except for systems used to detect financial fraud; and to evaluate and price life and health insurance for natural persons. It also covers AI systems intended to evaluate and classify emergency calls or to dispatch, or establish priority in dispatching, emergency first response services. Each use is tested on intended purpose, and the fraud exception and the life and health insurance limitation must be read exactly.
What public benefits uses does point 5 cover?
Point 5 covers systems used by or on behalf of public authorities to evaluate eligibility for essential public assistance benefits and services, including healthcare, and to grant, reduce, revoke or reclaim those benefits and services. The concern is consequential access to support that people depend on: a system that decides who receives a benefit, how much they receive, or whether support is withdrawn can affect health, housing and subsistence. A general information tool that explains benefits, without evaluating eligibility or deciding a grant, follows a different analysis. The record should name the benefit or service, the authority on whose behalf the system acts, the eligibility criteria, and the decision the system influences, because those facts determine whether the point is engaged.
What credit uses does point 5 cover?
Point 5 covers systems intended to evaluate the creditworthiness of natural persons or to establish their credit score, with an exception for AI systems used for the purpose of detecting financial fraud. Creditworthiness evaluation affects access to loans, housing and other essential financial services, and a score can follow a person across many decisions. The fraud exception is purpose-specific: a system used to detect fraud is not within the point on that ground, but a system that both detects fraud and evaluates creditworthiness must be assessed for the credit function separately. The record should separate the functions, describe the score, the data used, and the decisions informed, and should not treat a fraud label as covering a credit evaluation.
For the broader regulatory framework, see high-risk AI systems.
What insurance uses does point 5 cover?
Point 5 covers systems intended to evaluate and price life and health insurance for natural persons. The limitation to life and health insurance matters: other insurance lines, such as motor or property insurance, are not within this point on that ground, although other Annex III points, data protection law and insurance regulation can still apply. Life and health insurance pricing affects access to protection against the most consequential risks, and the evaluation can use health and other sensitive information. The record should name the insurance line, the pricing or evaluation function, the data used, and the safeguards for sensitive data. A system that only administers policies, without evaluating or pricing, follows a different analysis.
What emergency services uses does point 5 cover?
Point 5 also covers systems intended to evaluate and classify emergency calls by natural persons, or to be used to dispatch, or to establish priority in dispatching, emergency first response services, including police, firefighters and medical aid, as well as emergency patient triage systems. The concern is time-critical, life-affecting prioritisation: a misclassification can delay help. A system that transcribes a call without classifying or prioritising, or that provides general information, follows a different analysis. The record should describe the classification or dispatch function, the priority logic, the human oversight, and the failure safeguards, because the consequences of error in this context are immediate and severe.
How should the Article 6(3) filter be approached here?
The Article 6(3) filter can be considered for point 5 systems, but its conditions are demanding in this context. A benefits, credit, insurance or emergency system that materially influences an outcome, or that profiles natural persons, will commonly fail the filter, because profiling always prevents reliance on it and material influence defeats the no significant risk test. A narrow procedural task, such as formatting a call transcript without classifying it, is a different function from the listed use, and should be recorded as a separate function rather than as a filter argument for the listed use. Where the filter is relied on, the documentation and registration duties apply, and the reasoning must address profiling and material influence directly.
Continue with Article 6 classification routes.
What is not decided by this guide?
This guide explains point 5 in outline. It does not decide whether any particular benefits, credit, insurance or emergency system is high-risk, whether the fraud exception applies, or whether the Article 6(3) filter is available. Those questions need the function, the data, the decision affected, and the human review design, read against the consolidated Annex III wording. Systems in this point affect access to essential support and emergency help, so borderline cases should be confirmed with a competent authority or legal adviser before deployment.
Frequently asked questions
Is fraud detection in credit high-risk under point 5?
Point 5 excepts AI systems used for the purpose of detecting financial fraud from the creditworthiness group. A system that also evaluates creditworthiness must be assessed for that function separately, and other provisions can still apply.
Does point 5 cover all insurance pricing?
No. It covers evaluating and pricing life and health insurance for natural persons. Other insurance lines are not within this point on that ground, although sectoral regulation, data protection law and other Annex III points can still be relevant.
Is a triage chatbot in an emergency service high-risk?
It can be, where it evaluates and classifies emergency calls or dispatches or prioritises first response, or where it performs emergency patient triage. The function, not the chatbot label, decides, and the Article 6(3) filter must be tested on its strict conditions.
Continue through this AI Act cluster
This article is an informational overview, not a legal determination. Check the current legal text and official implementation guidance for your system and role.
Consolidated Regulation (EU) 2024/1689 ↗AI Act Service Desk timeline ↗European Commission regulatory framework for AI ↗