Key takeaways
- Classification depends on the system's intended purpose, not on the technology. A chatbot is not high-risk by default, but the same model used to screen job applicants is.
- There are two routes to high-risk: AI that is a safety component of a product under Annex I law, and AI used in one of the eight areas of Annex III.
- An Annex III system that profiles people is always high-risk. Others may qualify for an Article 6(3) exception, but the provider must document that assessment before launch and still register the system.
- High-risk duties for Annex III systems apply from 2 December 2027 under Regulation (EU) 2026/1744, and from 2 August 2028 for AI in regulated products. Check EUR-Lex before you plan around these dates.
- Small businesses are not exempt from classification. Fines for SMEs are capped at the lower of the fixed amount and the turnover percentage (Article 99).
How do you know if your AI system is high-risk?
Ask two questions. First, is the system a safety component of a product covered by the EU laws listed in Annex I, or is it such a product, and does that product need a third-party conformity assessment? Second, does it serve a purpose listed in Annex III? If the answer to either is yes, the system is high-risk (Article 6(1) and 6(2)), unless the Article 6(3) exception applies to an Annex III system. If both answers are no, the system is not high-risk, but Article 50 transparency and the GDPR can still apply.
Most business software falls outside both routes. A support chatbot, a meeting transcriber or a document search tool is usually not on the Annex III list. The use decides, so the same language model can be low risk in one product and high-risk in another. Since Regulation (EU) 2026/1744, AI used only for non-safety purposes such as user assistance, performance optimisation or convenience does not count as a safety component, unless its failure would endanger health or safety (Article 6(1a) and 6(1b)).
The eight Annex III areas in plain terms
| Area | Covered uses, in summary |
|---|---|
| 1. Biometrics | Remote biometric identification, biometric categorisation by sensitive traits and emotion recognition, where the use is not already banned. |
| 2. Critical infrastructure | Safety components in digital infrastructure, road traffic and the supply of water, gas, heating and electricity. |
| 3. Education and training | Admission decisions, assessing learning outcomes, deciding the level of education a person can access and detecting cheating in tests. |
| 4. Employment and worker management | Recruitment and selection (targeted job ads, application filtering, candidate evaluation), decisions on promotion or termination, task allocation and performance monitoring. |
| 5. Essential services | Eligibility for public benefits, credit scoring, risk assessment and pricing for life and health insurance, and emergency call triage. |
| 6. Law enforcement | Systems used by or for law enforcement authorities, such as risk assessment of individuals and evaluation of evidence. |
| 7. Migration, asylum and border control | Systems used by or for public authorities in these areas, such as risk assessment and the examination of applications. |
| 8. Justice and democratic processes | Systems that help a judicial authority research and apply the law, and systems meant to influence the outcome of an election or voting behaviour. |
This table summarises the Annex. Each point has its own conditions, so read the exact wording of the point that matches your system.
Can you claim an exception under Article 6(3)?
An Annex III system is not high-risk if it does not pose a significant risk of harm to health, safety or fundamental rights, including by not materially influencing the outcome of decision making, and at least one condition applies. The system performs a narrow procedural task, improves the result of a human activity already completed, detects decision patterns or deviations without replacing or influencing the human assessment, or performs a preparatory task for an assessment.
The exception never applies if the system profiles natural persons. Recruitment and credit scoring tools usually profile, so expect them to stay high-risk. If you rely on the exception, you must document your assessment before the system goes on the market and give it to the authorities on request (Article 6(4)). You must also register the system in the EU database (Article 49(2)). Regulation (EU) 2026/1744 streamlined the information that registration requires.
What changes once your system is high-risk?
Providers must run a risk management system, govern data quality, keep technical documentation and logs, design for human oversight, accuracy and robustness, and complete a conformity assessment (Articles 9 to 17 and 43). Deployers must follow the provider's instructions, assign trained people to oversee the system, keep the logs they control and inform affected workers and people (Article 26). Our EU AI Act checklists walk through these controls.
You become the provider if you put your own name on a high-risk system, modify it substantially or change a system's purpose so that it becomes high-risk (Article 25). SMEs get proportionate treatment, such as the lower fine cap in Article 99, but no exemption from classification.
Where do the Commission's guidelines stand?
On 19 May 2026 the Commission published draft guidelines on classifying high-risk systems under Article 6(5), with practical examples of systems that should and should not be classed as high-risk. They remain a draft for stakeholder feedback, and the Commission says the examples are not exhaustive. Treat them as an indication of its thinking and check for the final version.
Classify your system in four steps
- Write the intended purpose in one sentence: what the system does, who it affects and which decision it informs.
- Check Annex I. Is the system part of a regulated product, such as a toy, lift or medical device, that needs third-party assessment?
- Check the eight Annex III areas against your purpose. Compare the wording of the point, not only the heading.
- If you match Annex III, test whether Article 6(3) applies and whether the system profiles people. Record your reasoning and the date, whichever way you decide.
Send us a one-sentence description of your AI system, who it affects and which decision it informs. We reply within 4 business hours and can review whether it is likely to be high-risk and what compliance work a build would involve.
Check my system's risk levelFAQ
Is a chatbot high-risk under the AI Act?
Not by default. A customer-service chatbot is not listed in Annex III, but it must meet the Article 50 transparency rules. The same chatbot becomes high-risk if it performs an Annex III function, for example screening job applicants.
Is my AI system high-risk if I only buy it from a vendor?
The classification follows the system's purpose, whoever built it. If you use a high-risk system, you owe deployer duties under Article 26. You become the provider if you put your name on it, substantially modify it or change its purpose to a high-risk one (Article 25).
When do the high-risk rules start to apply?
For Annex III systems, from 2 December 2027, and for AI in Annex I products, from 2 August 2028, under Regulation (EU) 2026/1744. Confirm the dates on EUR-Lex before you plan.
Who decides whether my system is high-risk?
You do, as the provider, based on the intended purpose. Market surveillance authorities can review a classification that treats an Annex III system as not high-risk (Article 80), so document your reasoning.
Sources
- Regulation (EU) 2024/1689 (AI Act), Articles 6, 9 to 17, 25, 26, 43, 49, 50, 80 and 99 and Annexes I and III, Official Journal of the EU
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), Official Journal of the EU, 24 July 2026
- European Commission, Draft guidelines on the classification of high-risk AI systems, 19 May 2026
About the author
Artem Synytskyi · Co-founder · AI
Artem runs AI at ESADEX, building LLM, RAG and speech-to-text systems for regulated environments.