AI Act Guide
  1. Home
  2. High-Risk AI Systems under the EU AI Act

High-risk AI systems

High-Risk AI Systems under the EU AI Act: Classification, Requirements and Deadlines

How the EU AI Act classifies high-risk AI systems (Annex I, 8 Annex III areas, Article 6(3)), what applies, and dates 2 December 2027 and 2 August 2028.

Last updated 24 October 2026 · By the AIActGuide.net Editorial Team · How we fact-check · Print this page

High-risk AI systems under the EU AI Act are AI systems that either act as safety components of products covered by the 20 acts of Annex I or serve one of the 25 use cases in the eight areas of Annex III. Both routes lead to Chapter III: the seven requirements of Articles 9 to 15, the provider and deployer obligations, conformity assessment and registration. Chapter III Sections 1 to 3 apply from 2 December 2027 for Annex III systems and from 2 August 2028 for Annex I systems (Article 113, third paragraph, point (c)).

High-risk is one tier of risk among several. The EU AI Act is Regulation (EU) 2024/1689, in force since 1 August 2024 and amended by Regulation (EU) 2026/1744 of 8 July 2026 (the Digital Omnibus on AI), in force since 27 July 2026, which moved both high-risk dates. This page explains how a system becomes high-risk, what then applies, for whom and from when, citing the act as amended.

Compiled by the AIActGuide.net Editorial Team from the Official Journal text; how we check every statement: methodology.

What Is a High-Risk AI System?

A high-risk AI system is classified under Article 6(1) or Article 6(2); the classification triggers the Chapter III duties and does not ban the system. In legal terms, a high-risk AI system is an AI system that is a safety component of, or itself, an Annex I product requiring third-party conformity assessment, or an AI system intended for an Annex III use case that the Article 6(3) exception does not remove. Article 6 holds paragraphs 1, 1a to 1c (inserted in 2026) and 2 to 8 in the current text of 119 articles and 14 annexes, and the high-risk tier sits between the bans and the transparency duties among the EU AI Act risk categories.

The Two Routes to High-Risk: Article 6(1) Products and Article 6(2) Use Cases

Two routes make an AI system high-risk: the Annex I product route in Article 6(1) and the Annex III use-case route in Article 6(2). The table compares both routes as they read since 27 July 2026.

Route Legal basis Trigger Covers Applies from
Product route Article 6(1), Annex I Safety component of, or itself, a product under Annex I legislation that must undergo third-party conformity assessment Section A: 11 acts (points 2 to 12); Section B: 9 acts (points 13 to 21) 2 August 2028 (Article 113, third paragraph, point (c)(ii))
Use-case route Article 6(2), Annex III Intended purpose matches an Annex III use case, unless Article 6(3) applies 8 areas, 25 use cases 2 December 2027 (Article 113, third paragraph, point (c)(i))

The product route depends on sector law: the AI Act itself does not list products, it points to Union harmonisation legislation, such as the medical device, toy and lift rules. The use-case route depends on the intended purpose the provider sets.

High-Risk Is Not Prohibited: How the Tier Differs from the Article 5 Bans

High-risk classification permits placing on the market and use under conditions, while Article 5 prohibits listed practices outright. Article 6(2) classifies a system, and Chapter III then attaches requirements and obligations; nothing in Chapter III forbids the system. That rule differs from the prohibited AI practices of Article 5, whose points (a), (b) and (c) to (h) apply since 2 February 2025 and whose new points (ba) and (bb) and paragraphs 1a and 1b apply from 2 December 2026, as the guide to prohibited AI practices sets out. Breaching a ban carries the highest fine tier of EUR 35 000 000 or 7 % of total worldwide annual turnover, whichever is higher, while high-risk duties sit one tier lower.

The High-Risk Section of This Guide: Five Pages and What Each Answers

Five pages of this guide go deeper: the Article 6 rules, the Article 6(3) exception, Annex I, Annex III and the Commission guidelines.

Page What it answers Key provision
Article 6 classification rules Is a given AI system high-risk? Article 6(1) to (8)
Article 6(3) derogation When is an Annex III system not high-risk? Article 6(3), 6(4), 49(2)
Annex I Which product laws count for the product route? Annex I, Article 2(2)
Annex III high-risk use cases Which eight areas and 25 use cases are listed? Annex III, Article 7
Guidelines on the classification of high-risk AI systems What is the status of the Article 6(5) guidelines? Article 6(5)

The High-Risk Decision Logic in Four Steps

Four steps classify an AI system under Article 6: the Annex I product test, the Annex III use-case test, the Article 6(3) exception and the documentation. The steps give an orientation, not a legal assessment.

Step 1: The Product Test under Article 6(1)

An AI system is high-risk under Article 6(1) when it is a safety component of a product, or is itself a product, covered by Annex I legislation and that product must undergo third-party conformity assessment. Both conditions, points (a) and (b), are cumulative. Since 27 July 2026, Article 6(1a) states that AI systems "solely used for non-safety related aspects of user assistance, performance optimisation, service efficiency, automation or convenience or quality control shall not qualify as safety components", Article 6(1b) makes any system whose failure or malfunctioning would endanger health and safety a safety component, and Article 6(1c) adds that a third-party assessment required solely for other risks, such as radio spectrum or electromagnetic interference, does not meet point (b). For products under Annex I Section B, only Article 6(1), Article 60a and Articles 102 to 112 apply (Article 2(2) as replaced), and machinery moved there as point 21, Regulation (EU) 2023/1230, when Section A point 1 was deleted.

Step 2: The Use-Case Test under Article 6(2)

An AI system is high-risk under Article 6(2) when its intended purpose matches one of the 25 use cases in the eight Annex III areas. Annex III is unchanged by Regulation (EU) 2026/1744, and only a delegated act under Article 7(1) can add or modify use cases within the eight areas, using the 11 criteria of Article 7(2), points (a) to (k). The draft guidelines of 19 May 2026 state that systems assessing only legal persons fall outside the use cases that concern natural persons. The Commission's Article 112(1) report COM(2026) 234 of 20 May 2026 covers Article 5 and Annex III.

Step 3: The No-Significant-Risk Exception and the Profiling Override

An Annex III system is not high-risk when it poses no significant risk of harm and meets at least one of four alternative conditions in Article 6(3). Any one of the four conditions of the Article 6(3) derogation suffices:

  • Narrow procedural task (point (a)), such as turning unstructured data into structured data.
  • Improvement of a completed human activity (point (b)).
  • Pattern detection (point (c)): detecting decision-making patterns or deviations from prior patterns, without replacing or influencing the previously completed human assessment without proper human review.
  • Preparatory task (point (d)) for an assessment relevant to an Annex III use case.

An Annex III system that performs profiling of natural persons, as defined in Article 4(4) of Regulation (EU) 2016/679 through Article 3(52), is always high-risk, whatever condition it meets.

Step 4: Documenting a Not-High-Risk Assessment

A provider relying on Article 6(3) documents its assessment before placing the system on the market and registers it under Article 49(2). Article 6(4) also obliges the provider to hand that documentation to national competent authorities on request. Since 27 July 2026, the registration no longer includes the short summary of grounds or the list of Member States, because Regulation (EU) 2026/1744 deleted Annex VIII Section B points 7 and 9; under Article 80, the market surveillance authority can still evaluate the classification and fine a provider that misclassified a system to circumvent the requirements.

Annex III Areas at a Glance: Conformity Route, FRIA and Registration

Annex III lists 25 high-risk use cases in eight areas, each with its own conformity route, FRIA rule and registration place. In the table, "public bodies" means deployers that are bodies governed by public law or private entities providing public services.

Point Area Use cases Conformity route FRIA (Article 27) Registration (Article 49)
1 Biometrics 3 Annex VI or Annex VII (Article 43(1)) Public bodies EU database; secure non-public section in law enforcement, migration, asylum and border control (49(4))
2 Critical infrastructure 1 Annex VI None National level (49(5))
3 Education and vocational training 4 Annex VI Public bodies EU database
4 Employment, workers' management 2 Annex VI Public bodies EU database
5 Essential services, private and public 4 Annex VI Always for 5(b) and 5(c); otherwise public bodies EU database
6 Law enforcement 5 Annex VI Public bodies Secure non-public section (49(4))
7 Migration, asylum and border control 4 Annex VI Public bodies Secure non-public section (49(4))
8 Administration of justice and democratic processes 2 Annex VI Public bodies EU database

Annex III of Regulation (EU) 2024/1689, unchanged by Regulation (EU) 2026/1744; Articles 27(1), 43 and 49.

For point 1 systems of law-enforcement, immigration or asylum authorities or EU institutions, the market surveillance authority acts as the notified body (Article 43(1), third subparagraph). Points 5(b) and 5(c) cover creditworthiness and credit scoring, and risk assessment and pricing in life and health insurance.

What a High-Risk AI System Must Meet: Articles 8 to 15 at a Glance

Seven requirements in Articles 9 to 15 apply to high-risk AI systems under Annex III and Annex I Section A, met in the light of the intended purpose and the generally acknowledged state of the art (Article 8(1)):

Article Requirement Core duty
9 Risk management system Continuous iterative process planned and run throughout the entire lifecycle (9(2))
10 Data and data governance Training, validation and testing data relevant, sufficiently representative and, to the best extent possible, free of errors and complete (10(3))
11 Technical documentation Drawn up before placing on the market or putting into service and kept up to date (11(1))
12 Record-keeping Automatic recording of events (logs) (12(1))
13 Transparency and provision of information to deployers Instructions for use that are concise, complete, correct and clear (13(2))
14 Human oversight Ability to intervene or interrupt through a stop button or similar procedure (14(4)(e))
15 Accuracy, robustness and cybersecurity Accuracy levels and metrics declared in the instructions for use (15(3))

Bias detection with special categories of personal data moved to Article 4a on 27 July 2026, when Article 10(5) was deleted, and for remote biometric identification under Annex III point 1(a), Article 14(5) requires separate verification by at least two natural persons. A provider can rely on the presumption of conformity of harmonised standards (Article 40(1)), No common specifications have been adopted under Article 41 in the Cellar record as of 23 October 2026. Commission guidelines under Article 96(1)(g), due by 1 August 2027, explain how Articles 8(2), 9(10) and 17(3) apply alongside Annex I Section A legislation, and the guide to requirements for high-risk AI systems explains each article.

Provider, Deployer and Other Operator Duties for High-Risk AI Systems

Providers carry most high-risk duties (Article 16), deployers the use-side duties (Article 26), and importers, distributors and authorised representatives check compliance.

Article 3(8) names six operators, and the provider vs deployer roles can shift during the life of a system. A distributor, importer, deployer or third party becomes the provider under Article 25(1) when it puts its name or trademark on the system, makes a substantial modification, or changes the intended purpose so that the system becomes high-risk.

Operator Main article Core duties Retention Article 99(4) point
Provider Article 16 (12 points, (a) to (l)) Section 2 compliance, quality management system (Article 17, at least 13 aspects), conformity assessment, EU declaration, CE marking, registration, corrective action Documentation 10 years (18(1)); logs at least 6 months (19(1)) (a)
Deployer Article 26 (12 paragraphs); Article 27 for some Use per instructions, human oversight, input data, monitoring, worker information, FRIA where required Logs at least 6 months (26(6)) (e)
Importer Article 23 Check conformity assessment, documentation, CE marking and authorised representative before placing 10 years (23(5)) (c)
Distributor Article 24 Check CE marking, EU declaration and instructions before making available None set in Article 24 (d)
Authorised representative Article 22 Verify declaration and documentation, keep documents, cooperate with authorities 10 years (22(3)(b)) (b)
Product manufacturer Article 25(3) Annex I Section A product placed under its name: treated as the provider As provider As provider

Deployers assign human oversight to natural persons with the necessary competence, training and authority, keep the logs under their control, and inform workers' representatives and affected workers before using a system at the workplace (Article 26(7)). Public-authority deployers register their use (Article 26(8)), and a request to authorise post-remote biometric identification is made no later than 48 hours (Article 26(10)).

Bodies governed by public law, private entities providing public services and deployers of Annex III points 5(b) and 5(c) carry out a fundamental rights impact assessment before first use, covering six elements, points (a) to (f), and notify the results to the market surveillance authority (Article 27). The full map of EU AI Act obligations answers the wider question of what the act requires of each operator, and the page on the fundamental rights impact assessment covers Article 27 step by step.

Providers also run post-market monitoring, with Commission guidance and a voluntary template due by 2 September 2027 (Article 72(3)), and report serious incidents within 15 days, 2 days or 10 days depending on the case (Article 73).

Conformity Routes, Certificates and EU Database Entry for High-Risk AI Systems

Article 43 sets three conformity routes for high-risk AI systems. The guide to AI Act conformity assessment explains them for every risk route, and they differ by area and product type:

  • Annex III point 1 (biometrics): internal control under Annex VI only where harmonised standards or common specifications are applied in full; otherwise the Annex VII procedure with a notified body (Article 43(1)).
  • Annex III points 2 to 8: internal control under Annex VI, without a notified body (Article 43(2)).
  • Annex I Section A products: the conformity procedure of the sector legislation, which also checks the Section 2 requirements and the Article 17 quality management system (Article 43(3)).

Notified-body certificates are valid for at most five years for Annex I systems and four years for Annex III systems (Article 44(2)).

Providers register Annex III systems, except point 2, in the EU database before placing them on the market (Article 49(1)), public-authority deployers register their use (Article 49(3)), point 2 systems are registered at national level (Article 49(5)), and Annex I systems are not registered under Article 49. The page on EU database registration sets out the Annex VIII data items.

EN 18286:2026, the quality management system standard cited by Commission Implementing Decision (EU) 2026/2274 (OJ of 14 October 2026), is the only harmonised standard cited under the AI Act as of 24 October 2026. No notified body is listed on NANDO under the AI Act as of 24 October 2026. Notified bodies of the Annex I Section A sector laws apply for designation under the AI Act by 28 January 2028 (Article 43(3)).

When the High-Risk Rules Apply: Application Dates after the Digital Omnibus

Chapter III Sections 1 to 3 apply from 2 December 2027 to Annex III systems and from 2 August 2028 to Annex I systems (Article 113, third paragraph, point (c), as replaced by Regulation (EU) 2026/1744). Article 6(5) on the guidelines is excepted from the deferral, and Section 4 on notified bodies has applied since 2 August 2025. The AI Act timeline lists every other application date of the act.

Date Provision What applies Status on 24 October 2026
2 August 2025 Article 113, third paragraph, point (b) Chapter III Section 4 (Articles 28 to 39): notifying authorities and notified bodies Applies
2 August 2026 Article 113, second paragraph General date of application, including Article 50 transparency Applies
2 December 2027 Article 113, third paragraph, point (c)(i) Sections 1 to 3 for Annex III systems (AI Act 2 December 2027); before 27 July 2026: 2 August 2026 Upcoming
28 January 2028 Article 43(3) Annex I Section A notified bodies apply for AI Act designation Upcoming
2 August 2028 Article 113, third paragraph, point (c)(ii); Article 112(2) Sections 1 to 3 for Annex I systems (AI Act 2 August 2028); before 27 July 2026: 2 August 2027; first Article 112(2) evaluation Upcoming
2 August 2030 Article 111(2) High-risk systems of public authorities already on the market comply Upcoming

The dates moved by 16 months for Annex III and 12 months for Annex I because standards, common specifications and guidance arrived late and national competent authorities were not yet in place (recital 40 of Regulation (EU) 2026/1744).

The high-risk rules were delayed once, by an adopted regulation and not by a proposal. The Digital Omnibus on AI moved the high-risk dates when Regulation (EU) 2026/1744 entered into force on 27 July 2026.

The Commission FAQ of 20 October 2026 states: "The rules for high-risk AI systems will apply starting 2 December 2027."

Article 111(2) covers high-risk systems placed on the market before "the date of application of Chapter III referred to in Article 113" only if their design changes significantly from that date. The page on AI systems already on the market explains this transitional rule.

Penalties for Breaching High-Risk Duties (Article 99(4))

Breaches of high-risk operator duties carry fines of up to EUR 15 000 000 or 3 % of total worldwide annual turnover, whichever is higher (Article 99(4)). The list covers Articles 16, 22, 23, 24, 26 and 50, and the duties of notified bodies. Since 27 July 2026, point (da) adds Article 25(2) and (4).

A provider that misclassifies a system to circumvent the requirements is fined under Article 99 (Article 80(7)). Fines on SMEs, including start-ups, and on small mid-caps are capped at the lower of the two amounts in the EUR 15 million tier (Article 99(6) and (6a)).

Tier Provisions Maximum
Article 99(4) Operator duties (Articles 16, 22 to 24, 25(2) and (4), 26, 50), notified bodies EUR 15 000 000 or 3 % of total worldwide annual turnover, whichever is higher
Article 99(5) Incorrect, incomplete or misleading information to notified bodies or authorities EUR 7 500 000 or 1 % of total worldwide annual turnover, whichever is higher
Article 99(3), for contrast Prohibited practices under Article 5 EUR 35 000 000 or 7 % of total worldwide annual turnover, whichever is higher

The lower-amount cap covers all three tiers for SMEs and start-ups (Article 99(6)), but only the Article 99(4) and (5) tiers for small mid-caps (Article 99(6a)).

Article 99(4) does not list Article 27, so penalties for a missing fundamental rights impact assessment follow national rules under Article 99(1). The guide to EU AI Act penalties sets out the full regime, including fines for general-purpose AI models.

How to Prepare for the High-Risk Rules: Compliance Steps and Checklist

Five steps prepare a provider or deployer for 2 December 2027, when Chapter III Sections 1 to 3 apply to Annex III systems: inventory, classification, role assignment, gap analysis and conformity planning. The same steps apply to Annex I products, with 2 August 2028 as the date, in this order:

  1. List every AI system in use or development, with its intended purpose.
  2. Classify each system against Annex I, Annex III and Article 6(3), and document any not-high-risk assessment.
  3. Assign the operator role per system: provider, deployer, importer, distributor, authorised representative or product manufacturer.
  4. Compare current controls with Articles 9 to 15, Article 17 and Article 26, including the FRIA duty of Article 27.
  5. Plan the conformity route, registration and the evidence the technical documentation needs.

The inventory and the classification come first because every later duty follows from them. A step-by-step programme for both roles is set out in the guide to EU AI Act compliance, written for providers and deployers alike. Article 4 AI literacy already applies since 2 February 2025: since 27 July 2026, providers and deployers "shall take measures to support the development of AI literacy".

A gap analysis starts from the seven requirements and the operator table above.

Each duty can be traced to its article and date. The EU AI Act checklist sorts the obligations by role and risk class, with every deadline.

Article 50, Chapter V and the sector product laws apply alongside the high-risk rules; the Article 6(5) guidelines exist only as a draft of 19 May 2026.

What Is the Status of the Article 6(5) Guidelines?

The Commission guidelines on high-risk classification exist only as a draft of 19 May 2026, although Article 6(5) required them by 2 February 2026.

The register of EU AI Act guidelines lists the draft, in three parts of 6, 13 and 148 pages, with all other guidelines. The Commission announced on 19 May 2026 that the final guidelines will be adopted by the end of 2026.

Can a System Built on a General-Purpose Model Be High-Risk?

General-purpose AI models under Chapter V are not high-risk systems themselves, but an AI system built on one can be high-risk.

The obligations for general-purpose AI models sit in Articles 51 to 56. Under Article 25(1), point (c), whoever changes the intended purpose of a general-purpose AI system so that it becomes high-risk becomes its provider.

Do Article 50 Duties Apply on Top of the High-Risk Rules?

Article 50 applies since 2 August 2026, on top of any high-risk duties.

The transparency obligations under Article 50 cover chatbots, synthetic content and deepfakes. Article 13 is a separate high-risk requirement on information to deployers.

How Do Sector Product Laws Interact with the High-Risk Rules?

Sector product laws stay in force, and for Annex I Section A products the AI Act requirements are checked in the sector conformity procedure (Article 43(3)). The Machinery Regulation, Regulation (EU) 2023/1230, applies from 14 January 2027. Article 42(3) deems systems meeting Article 12(1) of the Cyber Resilience Act to meet the Article 15 cybersecurity requirement.

Do Non-EU Providers of High-Risk AI Systems Have to Comply?

Non-EU providers carry the same high-risk duties when they place systems on the Union market.

The guide to the EU AI Act for non-EU companies explains how the act reaches them. Article 22 requires them to appoint an authorised representative established in the Union by written mandate before making a system available.

Which Authorities Supervise High-Risk AI Systems?

Market surveillance authorities supervise high-risk AI systems and evaluate Article 6(3) assessments (Article 80).

The national competent authorities of the 27 Member States are listed by country. The AI Office, a Commission function, supervises general-purpose AI models.

Frequently Asked Questions about High-Risk AI Systems

What is considered a high-risk AI system?

An AI system that is a safety component of, or itself, an Annex I product needing third-party conformity assessment, or that serves one of the 25 use cases in the eight Annex III areas, unless Article 6(3) applies.

Are high-risk AI systems banned?

No. High-risk AI systems may be placed on the market and used once they meet Chapter III. Only the practices in Article 5 are prohibited: points (a), (b) and (c) to (h) since 2 February 2025, points (ba) and (bb) and Article 5(1a) and (1b) from 2 December 2026.

When do the high-risk rules apply?

Annex III systems from 2 December 2027 and Annex I systems from 2 August 2028, the dates on which Chapter III Sections 1 to 3 apply to them. Before 27 July 2026, the dates were 2 August 2026 and 2 August 2027.

Does every high-risk AI system need a notified body?

No: Annex III points 2 to 8 use internal control (Annex VI); point 1 can need a notified body (Annex VII); Annex I Section A products go through the third-party procedure their sector legislation requires (Articles 6(1)(b) and 43(3)).

Who checks whether a provider classified its system correctly?

The national market surveillance authority, under Article 80. It evaluates systems classified as not high-risk under Article 6(3) and can require the provider to bring them into compliance.

Do deployers of high-risk AI systems have obligations?

Yes, under Article 26: use per the instructions, human oversight, log keeping for at least six months and worker information. Some deployers also carry out a FRIA under Article 27.

Sources