Is your AI system high risk? Answer and you will know.
For anyone who develops an AI system or puts one into use in their business. Under the European AI Regulation, we tell you whether it fits a high-risk use, whether any exception applies and what you have to do according to your role. We do not keep your answers.
- Who it is for
- For whoever develops an AI system or puts it into use in their company. If you only write to a chatbot to draft or translate, it is almost certainly not high-risk: what counts is what it is used for.
- What it looks at
- Whether the system falls under any of the uses in Annex III of the European AI Regulation, whether any of the exceptions in art. 6, paragraph 3, applies, and what you have to do depending on whether you are a provider or a deployer.
- What you take away
- A result in plain language, with the references to the law, the date from which it applies and a list of obligations. Nothing is stored.
This is not legal advice. It is guidance based on what you answer. If you only want the list of your team's AI tools, use the AI inventory. If you have already tested your system with promptfoo or garak, upload the result and we will explain it to you.
Examples
- A company uses a program that scores candidates and discards CVs. High risk (Annex III, point 4, letter a)
- A system proposes who should be promoted and its proposal weighs on the salary of the new post. High risk (Annex III, point 4, letter b)
- A town council uses a system to decide who receives a grant and how much. High risk (Annex III, point 5, letter a)
- A bank scores individuals' creditworthiness with its own model. High risk (Annex III, point 5, letter b)
- An exam-proctoring program flags students who cheat. High risk (Annex III, point 3, letter d)
- An insulin pump that doses on its own, inside a medical device. Possible high risk through the product (art. 6, paragraph 1, and Annex I)
- A team of people uses a chatbot to summarise emails and translate. Not high risk (is not in Annex III)
- A shop handles enquiries with a chatbot (people must be told that it is an AI: art. 50). Not high risk (is not in Annex III)
The examples are based on AESIA Guide 2, © 2026 Ministerio para la Transformación Digital y de la Función Pública, under a Creative Commons Attribution 4.0 International licence; we have summarised and changed them. Source of the data: Agencia Española de Supervisión de Inteligencia Artificial.
Harmonised standards and presumption of conformity
If you comply with a harmonised standard whose reference has been published in the Official Journal, you are presumed to meet the requirements of section 2 that the standard covers (art. 40, paragraph 1).
As of 31 October 2026 there is only one published reference for the AI Act: EN 18286:2026, on the quality management system of providers (Implementing Decision (EU) 2026/2274, Official Journal of 14 October 2026).
There are no harmonised standards published yet for the other requirements (risks, data, logs, cybersecurity...). Today you cannot rely on a presumption of conformity for them: use the guides and assess your system.
Note: Article 40(1) refers to the requirements of Section 2, and that standard is about the quality system (Art. 17). Read the Decision itself to see exactly what presumption it gives, and consult whoever advises you.
- EN 18286:2026: Artificial intelligence: quality management system for regulatory purposes under the EU Artificial Intelligence Act. Commission Implementing Decision (EU) 2026/2274 (opens in another tab), published on 14 October 2026.
Checked in the Official Journal of the European Union on 31 October 2026.
Which dates apply today
- 2 February 2025
- AI literacy (art. 4) and prohibited practices (art. 5).
- 2 August 2026
- General date of application of the Regulation (art. 113, second paragraph).
- 2 December 2027
- High risk under Annex III: requirements and obligations of providers and deployers (art. 113, third paragraph, point (c), sub-point (i)).
- 2 August 2028
- High risk through regulated products, under Annex I (art. 113, third paragraph, point (c), sub-point (ii)).
- 2 August 2030
- High-risk systems for public authorities that were already on the market before the date of application (art. 111, paragraph 2).
Frequently asked questions
Am I a provider or a deployer?
A provider is whoever develops an AI system, or has it developed, and offers it or puts it into service under their own name or brand (Art. 3(3)). A deployer is whoever uses it under their own authority, except for personal non-professional use (Art. 3(4)). You can be both. And if you put your brand on someone else's high-risk system, or change what a system is for so that it becomes high-risk, you become a provider (Art. 25).
When do the high-risk obligations apply?
For the Annex III uses, from 2 December 2027; for the Annex I products, from 2 August 2028 (Art. 113, following Regulation (EU) 2026/1744). Systems already on the market before then are only covered if they later change significantly in design, and those of public authorities have until 2 August 2030 (Art. 111(2)).
What is the exception in art. 6, paragraph 3?
An Annex III system is not considered high-risk if it does not pose a significant risk to people's health, safety or rights, including because it does not materially influence the outcome of the decision. This is deemed to be met if it performs only a limited procedural task, improves something a person has already done, detects patterns without replacing a person or prepares an assessment. It never applies if it profiles people.
If I think my system is not high-risk, what do I do?
If you are the provider, you have to document your assessment before placing it on the market or putting it into service and register it in the EU database; the authority can ask you for that documentation (art. 6, paragraph 4, and art. 49, paragraph 2). If you are the one deploying it, ask the provider what it is based on.
Are there harmonised standards that give me a “presumption of conformity”?
Only one reference, on the quality management system, has been published as of 31 October 2026. See the harmonised standards section of this page.
Where can I see the official guidelines?
Article 6(5) asked the Commission for guidelines with examples by 2 February 2026 at the latest. As of 31 October 2026 there is only a draft (19 May 2026), and the Commission has an online checker in beta. AESIA has published practical guides. The links are below.
Is what I answer stored?
No. It is calculated on the spot and returned to you on the same page. We do not save your answers or send them to anyone.
Does this replace an adviser?
No. It is guidance based on what you answer. It is not legal advice: if in doubt, consult whoever advises you.
Official guides and checkers
- Commission draft guidelines on the classification of high-risk systems (draft, in English) (opens in another tab)
- The Commission's Compliance Checker (beta version) (opens in another tab)
- Practical guides from AESIA (the Spanish Agency for the Supervision of Artificial Intelligence) (opens in another tab)
Official sources
- Regulation (EU) 2024/1689, the AI Act: Arts. 2, 3, 6, 16, 25, 26, 27, 40, 43, 49 and 113 and Annexes I and III (opens in another tab)
- AI Act, consolidated version of 27 July 2026 (opens in another tab)
- Regulation (EU) 2026/1744, the Digital Omnibus on AI (opens in another tab)
- Commission Implementing Decision (EU) 2026/2274: reference of the harmonised standard EN 18286:2026 (opens in another tab)
Read on 31 October 2026 in the text of the Official Journal of the European Union. This is not legal advice: it is guidance based on what you answered. If in doubt, check with whoever advises you.