High-risk AI systems: what counts as high-risk and what to do about it.
The EU AI Act places most of its obligations on high-risk AI systems. Which systems count depends on the purpose, not the technology. This guide shows the Annex III areas, the exemptions, the duties for providers and deployers, and the deadlines.
As of 28 October 2026 · not legal advice
Key points
- An AI system is high-risk if it is a safety component of a regulated product (Annex I) or is used in one of the eight Annex III areas, such as employment, education or creditworthiness.
- By way of exception, an Annex III system is not high-risk if it poses no significant risk, for example when they perform a narrow procedural task (Art. 6(3)). If a system profiles natural persons, the exemption never applies.
- Providers carry most duties: risk management, documentation, logging, human oversight and conformity assessment. Deployers use the system as instructed, ensure oversight and keep logs under their control for at least six months.
- Following the Digital Omnibus, the duties apply from 2 December 2027 (Annex III) and 2 August 2028 (Annex I).
What “high-risk” means in the EU AI Act.
The EU AI Act sorts AI systems by risk. Prohibited practices are banned, certain systems carry transparency duties, and high-risk AI systems face the strictest requirements. A system becomes high-risk in two ways:
Route 1: products (Annex I)
The AI system is a safety component of a product, or a product itself, that needs third-party assessment under EU product law, such as machinery or medical devices.
Route 2: areas (Annex III)
The AI system is used for a purpose in one of the eight areas in Annex III, such as selecting applicants or assessing creditworthiness.
Purpose, not technology
Whether a system is high-risk depends on what it is intended for. The same language model can be high-risk in one use and not in another.
The duties for high-risk systems apply from 2 December 2027 (Annex III) and from 2 August 2028 (Annex I).
EU AI Act: current status →High-risk AI systems: the eight Annex III areas.
| Area | Examples of listed purposes |
|---|---|
| 1. Biometrics | remote biometric identification, categorisation by sensitive attributes, emotion recognition (where permitted) |
| 2. Critical infrastructure | safety components for digital infrastructure, road traffic, water, gas, heating, electricity |
| 3. Education | access and admission, evaluating learning outcomes, monitoring during tests |
| 4. Employment | selecting applicants, decisions on promotion or termination, task allocation, performance evaluation |
| 5. Essential services | eligibility for public benefits, creditworthiness, risk and pricing in life and health insurance, emergency calls |
| 6. Law enforcement | certain assessments and analyses by authorities (where permitted) |
| 7. Migration and borders | risk assessments, examining asylum, visa or residence applications (where permitted) |
| 8. Justice and democracy | assisting judicial authorities in research and interpretation, influencing elections |
The table summarises Annex III. The exact wording is what counts; the European Commission can amend the list.
Duties for providers and deployers.
The EU AI Act distinguishes who develops a system and places it on the market (provider) from who uses it under their own authority (deployer). Most duties sit with the provider.
| Time | User | Action |
|---|---|---|
| 24.09. 09:14 | a.weber | question asked |
| 24.09. 08:52 | a.weber | source opened |
| 23.09. 17:30 | p.neumann | language model changed |
| 23.09. 11:05 | a.weber | retention changed |
| Duty | Provider | Deployer |
|---|---|---|
| Risk management and data governance | yes (Art. 9, 10) | keep input data relevant (Art. 26) |
| Technical documentation | yes (Art. 11) | – |
| Logging | enable automatic logs (Art. 12) | keep logs for at least six months (Art. 26) |
| Human oversight | enable it (Art. 14) | assign competent people (Art. 26) |
| Conformity assessment and registration | yes (Art. 43, 49) | public authorities: registration (Art. 26, 49) |
| Information | instructions for use (Art. 13) | inform workers’ representatives and affected people (Art. 26) |
A deployer who offers a high-risk system under their own name, substantially modifies it, or changes a system’s purpose so that it becomes high-risk is treated as a provider (Art. 25).
Is my AI assistant high-risk?
An AI assistant is not automatically high-risk just because it uses a language model. What matters is what you use it for. These four questions help with the assessment; record the result in writing.
1. Area
Does the purpose fall into one of the eight Annex III areas, such as selecting people or deciding on benefits?
2. Exemption
Does the system only perform a narrow procedural task, improve a completed human result, detect decision-making patterns or deviations without replacing or influencing a completed human assessment, or only prepare an assessment (Art. 6(3))?
3. Profiling
Does the system evaluate people based on personal characteristics? Then an Annex III system always stays high-risk.
4. Documentation
A provider who considers an Annex III system not high-risk must document that assessment before placing it on the market and register the system (Art. 6(4), Art. 49(2)).
This is general information, not legal advice. We do not classify RAGSuite or your use; check the specific purpose with your legal adviser.
What you can do now.
Map your AI use
Which AI systems does your company use, for what and with which data?
Check the purpose
For each system, clarify whether an Annex III area is involved and whether an exemption applies.
Clarify your role
Are you provider, deployer or both? The duties follow from that.
Oversight and logs
Name the people responsible and decide where logs are kept and for how long.
Involve your people
Inform workers’ representatives early and build AI literacy (Art. 4).
Questions about high-risk AI systems.
What is a high-risk AI system?
An AI system that is a safety component of a regulated product (Annex I) or is used for a purpose in one of the eight Annex III areas, such as employment, education, creditworthiness or access to public benefits. These systems face the strictest duties in the EU AI Act.
When do the high-risk duties apply?
Following the Digital Omnibus (Regulation (EU) 2026/1744), they apply to Annex III systems from 2 December 2027 and to systems in regulated products under Annex I from 2 August 2028. Other parts of the EU AI Act, such as the prohibitions, AI literacy and Art. 50, already apply.
What exemptions are there?
An Annex III system is not high-risk if it poses no significant risk to health, safety or fundamental rights and, for example, only performs a narrow procedural task or only prepares an assessment (Art. 6(3)). If it profiles people, it always stays high-risk. The provider must document the assessment and register the system.
What do deployers have to do?
Deployers use the system according to the instructions, assign competent people for oversight, make sure input data is relevant, monitor operation, keep logs under their control for at least six months, and inform workers’ representatives and affected people. Public authorities must also register the system. Public bodies and certain deployers, for example in credit and insurance, also carry out a fundamental-rights impact assessment (Art. 27).
Is an AI chatbot high-risk?
Not automatically. A chatbot that answers questions from documents is not listed as a purpose in Annex III. If it is used, for example, to select applicants or decide on benefits, it can be high-risk. Separately, the transparency duty under Art. 50 applies to chatbots. This is not legal advice.
Use your knowledge, keep the duties in view.
Stefan shows you how to run RAGSuite in your own network, with roles and a full audit log in Enterprise.