Commission draft guidance on high-risk AI classification

The Commission's third interpretive Communication on the EU AI Act addresses Article 6 high-risk classification. It was published in draft form on 19 May 2026 for stakeholder consultation closing 23 June 2026, in three companion documents: a general-principles annex, an Article 6(1)+Annex I product-safety annex, and an Article 6(2)+Annex III use-case annex. This page paraphrases the Commission's classification framework. The Annex III worked examples are on the companion page High-risk worked examples.

Status — DRAFT

Commission interpretive guidance, not binding law. The EU AI Act text and any Court of Justice of the European Union (CJEU) interpretation prevail in case of conflict.

This page reflects the draft Commission Guidelines published in May 2026. The consultation closes 23 June 2026 and the text may change before formal adoption. The May 2026 provisional AI Omnibus agreement also amends the underlying Article 6 timeline — see the EU AI Act overview.

Quick decision

• You want to know whether your system is high-risk → Article 6 has two statutory high-risk scenarios plus an Article 6(3) Annex III filter check: (1) Article 6(1) + Annex I Section A (NLF product-safety legislation); (2) Article 6(1) + Annex I Section B (other Union harmonisation legislation); (3) Article 6(2) + Annex III (eight use-case categories); plus (4) the Article 6(3) filter mechanism, which may allow an Annex III system to be treated as not high-risk if one of four alternative conditions (a)–(d) is met, the system does not materially influence decision-making, and the system does not perform profiling.
• Your system is in Annex III → it is high-risk by default. Check the Article 6(3) filter mechanism for a possible non-high-risk determination. Profiling is an absolute bar to the filter.
• Your system is a safety component of a regulated product → check whether the product is under one of the Annex I acts and whether the product is required to undergo third-party conformity assessment. Both conditions must hold.
• You determine your Annex III system is not high-risk under Article 6(3) → Article 6(4) documentation obligation + Article 49(2) EU-database registration apply. Maps to Modulos MRF-111.
• You're modifying or rebadging an existing system → Article 25(1) makes you a provider if you substantially modify a high-risk system, rebadge it, or modify the intended purpose of a non-high-risk system in a way that makes it high-risk.

TL;DR

• Article 6 sets two classification scenarios (paragraph (7) general principles): Article 6(1) product-safety route via Annex I; Article 6(2) standalone route via Annex III.
• The AI-system test (Article 3(1)) is the prerequisite — paragraphs (8)–(9) general principles. A system that is not an AI system cannot be high-risk. See Commission guidance on the AI-system definition.
• Intended purpose (Article 3(12)) is the controlling concept (paragraphs (10)–(13) general principles). Broad multi-purpose framing will be treated as encompassing high-risk uses; service-terms exclusions alone are insufficient where the overall presentation provides for high-risk uses.
• Article 6(3) provides four alternative conditions for the Annex III filter mechanism (a non-high-risk determination), but a profiling override always re-classifies the system as high-risk (third sub-paragraph of Article 6(3); paragraph (89) Annex III annex).
• Article 6(4) documentation + Article 49(2) EU-database registration apply to non-high-risk Annex III determinations.
• Timeline (post-Omnibus, pending adoption): Article 6(2) Annex III obligations apply from 2 December 2027 (originally 2 August 2026); Article 6(1) Annex I obligations apply from 2 August 2028 (originally 2 August 2027).

Primary source

Draft Commission Guidelines on the classification of high-risk AI systems under Article 6 of the EU AI Act for stakeholder consultation, May 2026. Three companion documents: general principles, Article 6(1) Annex I route, Article 6(2) Annex III route. © European Union. Draft guidelines issued pursuant to Article 6(5) AI Act; not yet adopted. Consultation closes 23 June 2026; text may change.

The Article 6 classification framework

Per Article 6 AI Act and the Commission's draft general principles at paragraph (7), Article 6 has two statutory high-risk scenarios; operationally, this page separates them into these checks:

Check	Classification effect	Authority
A — Article 6(1) + Annex I Section A	AI is a safety component of (or is) a product under NLF legislation requiring third-party conformity assessment	Article 6(1), Annex I Section A; New Legislative Framework
B — Article 6(1) + Annex I Section B	AI is a safety component of (or is) a product under other Union harmonisation legislation requiring third-party conformity assessment	Article 6(1), Annex I Section B
C — Article 6(2) + Annex III	AI's intended purpose is listed in one of the eight Annex III categories	Article 6(2), Annex III
D — Article 6(3) filter (Annex III only)	Annex III system may be treated as not high-risk if at least one condition (a)–(d) is met, the system does not materially influence the decision outcome, and it does not perform profiling	Article 6(3); paragraphs (86)–(90)

Checks A and B are about products with AI in them. Check C is about AI systems with high-risk intended purposes. Check D is the filter back-out from Check C only; it does not apply to Checks A or B.

Prerequisite — the AI-system test

Article 6 only applies to AI systems within the meaning of Article 3(1). A system that fails the Article 3(1) test (e.g. a basic data-processing system) is not within the scope of the AI Act at all and therefore cannot be high-risk. The Commission cross-references its definition guidance C(2025) 5053 at paragraph (9) general principles.

This sequencing matters operationally: a credit-scoring system that is mechanically a "linear regression" implementation does not escape Annex III(5)(b) unless it fails the Article 3(1) AI-system test on its specific architecture and functionality — and the Commission's definition guidance is clear that a high-stakes intended purpose alone does not flip a basic-data-processing system into the AI Act.

Intended purpose — the controlling concept

Per Article 3(12) AI Act and the Commission's draft general principles at paragraphs (10)–(13), 'intended purpose' is the use for which the system is intended by the provider, as specified in:

• the instructions for use;
• promotional or sales materials and statements;
• the technical documentation.

The Commission emphasises three operational consequences (paragraph (11)–(12)):

1. Clarity is the provider's responsibility. The functionalities and intended use should be unambiguous across all materials.
2. Multi-purpose framing is dangerous. If the provider presents the system as broadly applicable across contexts and functions, and does not consistently exclude high-risk uses, the intended purpose is deemed to encompass those high-risk uses.
3. Terms-of-service exclusions are insufficient on their own. Merely asserting in terms of service that a high-risk use is excluded does not avoid high-risk classification where the provider's "overall presentation, examples, or product positioning effectively provides for or promotes such uses" (paragraph (12)).

This is the single most operationally consequential paragraph of the draft for Scout-style queries. A general-purpose AI system marketed broadly, without clear and coherent limitations excluding high-risk uses, will be deemed Annex III-applicable where such uses are feasible and reasonably foreseeable given the system's functionalities and capabilities — contractual exclusions alone are insufficient.

Article 6(1) — Annex I product-safety route

Per the draft Annex I annex at paragraphs (15)–(27):

Two cumulative conditions

Article 6(1) imposes two cumulative conditions for high-risk classification:

1. The AI system is intended to be used as a safety component of a product, or is itself a product, regulated by Union harmonisation legislation listed in Annex I (the Annex I list is exhaustive — only legislative changes to Annex I expand or contract scope per paragraph (24)).
2. The product is required to undergo a third-party conformity assessment under that legislation (with a view to placing it on the market or putting it into service).

Both conditions must hold. A safety component of a product under Annex I legislation that does not require third-party conformity assessment is not high-risk under Article 6(1).

Section A versus Section B

Annex I has two sections:

• Section A — New Legislative Framework (NLF) legislation. Examples per paragraph (23): Machinery Regulation, Toy Safety Directive, Lifts Directive, Radio Equipment Directive, Pressure Equipment Directive, Recreational Craft Directive, Personal Protective Equipment Regulation, Appliances Burning Gaseous Fuels Regulation, Medical Devices Regulation (EU) 2017/745, In Vitro Diagnostic Medical Devices Regulation (EU) 2017/746.
• Section B — Other Union harmonisation legislation. Examples: motor-vehicle type-approval, civil aviation, marine equipment, agricultural and forestry vehicles.

The Commission's draft notes (paragraph (25)) that for safety-component classification, the AI Act focuses on health, safety and fundamental rights — public interests under the Annex I legislation that go beyond these (e.g. radio spectrum use, electromagnetic compatibility) are excluded from the Article 6(1) classification analysis.

Independent / embedded / remote AI

Per paragraph (22), Article 6(1) applies irrespective of whether the AI system is embedded within the product or placed on the market or put into service independently — for example, as a software update, an add-on, or a remote service. The form of supply does not change the classification analysis.

What is excluded

Per paragraph (21), not all AI systems that are components of regulated products are high-risk. Many consumer products will fall under Annex I legislation but will not meet one or more of the cumulative conditions — for example, smart home appliances that fall under product-safety legislation but do not require third-party conformity assessment will not be classified as high-risk under Article 6(1) (though may still face Article 6(2) classification on a separate basis).

Article 6(2) — Annex III use-case route

Per the draft Annex III annex (extensive treatment across paragraphs (66) onwards and per-sub-point sections 3.1 through 3.8):

The eight Annex III categories

Annex III lists eight areas of high-risk use cases (Article 6(2) reads each as a positive trigger):

1. Biometrics — 1(a) post-RBI for law enforcement evidence; 1(b) biometric categorisation in sensitive areas (with some exceptions); 1(c) emotion recognition.
2. Critical infrastructure — safety components of management/operation of critical digital infrastructure, road traffic, supply of water, gas, heating, electricity.
3. Education and vocational training — 3(a) access/admission; 3(b) evaluating learning outcomes; 3(c) assessing appropriate level of education; 3(d) monitoring prohibited behaviour during tests.
4. Employment, workers management, access to self-employment — 4(a) recruitment or selection; 4(b) decisions affecting work relationships, promotion, termination, monitoring, evaluation.
5. Access to and enjoyment of essential private services and essential public services and benefits — 5(a) eligibility for public assistance; 5(b) creditworthiness / credit scoring of natural persons (excluding fraud detection); 5(c) risk assessment and pricing in life and health insurance; 5(d) emergency-call evaluation / dispatching prioritisation.
6. Law enforcement — including 6(a) victim assessment; 6(b) polygraph-style evidence reliability; 6(c) crime detection; 6(d) profiling for criminal-investigation purposes; 6(e) profiling under Directive 2016/680; etc.
7. Migration, asylum and border control — 7(a) polygraph-style at borders; 7(b) risk assessment of migrants; 7(c) document authenticity assessment; 7(d) examining applications.
8. Administration of justice and democratic processes — 8(a) research and interpretation of facts and law by judicial authorities; 8(b) influencing elections.

Per the Commission's draft, intended-purpose framing is the controlling test for Article 6(2): the system must be intended to be used for one of the listed uses. A general-purpose system marketed broadly will be deemed Annex III-applicable wherever Annex III uses are feasible and reasonably foreseeable per paragraph (12) general principles.

Worked examples per sub-point are on the High-risk worked examples page.

Article 6(3) — the filter mechanism

Per the draft Annex III annex at section 2.7 and paragraphs (84)–(89):

Four alternative conditions

Article 6(3) provides that a provider of an Annex III system may exempt the system from high-risk classification if any one of the following conditions is met:

• (a) The AI system is intended to perform a narrow procedural task.
• (b) The AI system is intended to improve the result of a previously completed human activity.
• (c) The AI system is intended to detect decision-making patterns or deviations from prior decision-making patterns and is not meant to replace or influence the previously completed human assessment without proper human review.
• (d) The AI system is intended to perform a preparatory task to an assessment relevant for the purposes of the use cases listed in Annex III.

The conditions are alternative, not cumulative (paragraph (88) Annex III annex). One condition is enough, but the conditions must be interpreted narrowly because Article 6(3) is an exception from rules that protect fundamental rights; the system must not materially influence the outcome of decision-making, and paragraph (90) excludes reliance on the filter where a module forms part of a complex system whose combined intended purpose or joint outputs materially influence an individual decision.

Profiling override

Profiling override (third sub-paragraph of Article 6(3))

An AI system referred to in Annex III shall always be classified as high-risk where the system performs profiling.

Profiling is read by cross-reference to Article 4(4) GDPR, Article 3(4) Law Enforcement Directive, and Article 3(5) Regulation (EU) 2018/1725. Per the Commission's draft Annex III annex at paragraph (89) and section 2.7.2, the profiling override applies regardless of whether the system meets one of the (a)–(d) conditions.

This is the most-likely-missed clause of Article 6 — a system that meets a "narrow procedural task" condition but processes personal data in a way that meets the Article 4(4) GDPR profiling definition is high-risk.

Article 6(3) is filter only — not for Article 6(1)

Per paragraph (87) Annex III annex, the filter mechanism applies only to AI systems that would normally be classified as high-risk pursuant to Article 6(2). It does not apply to AI systems classified under Article 6(1) (the Annex I product-safety route). A safety component of a regulated product cannot exit high-risk classification through Article 6(3).

Article 6(4) documentation + Article 49(2) registration

Per Article 6(4) AI Act, a provider that determines an Annex III system is not high-risk under Article 6(3) must:

1. Document the assessment before the system is placed on the market or put into service.
2. Register the system in the EU database referred to in Article 71, via the Article 49(2) registration obligation.
3. Provide the documentation on request to market surveillance authorities.

The May 2026 provisional AI Omnibus agreement (pending formal adoption) reinstated this Annex III non-high-risk registration requirement; the Commission's original Omnibus proposal had removed it. See the EU AI Act overview.

Substantial modification + value-chain triggers

Per Article 25(1) AI Act and the Commission's draft general principles at paragraph (14), distributors, importers, deployers, and other third parties may become subject to provider obligations if they:

• (i) put their name or trademark on a high-risk AI system already placed on the market or put into service;
• (ii) make a substantial modification to a high-risk AI system already placed on the market or put into service, in such a way that it remains a high-risk AI system; or
• (iii) modify the intended purpose of an AI system (including a general-purpose AI system) that has not been classified as high-risk, in such a way that the AI system becomes high-risk under Article 6.

The Commission is preparing separate guidelines on Article 25 value-chain responsibilities. Until those are adopted, the operative reference is the Regulation text + paragraph (14) of the high-risk general-principles annex.

Post-Omnibus timeline

Per Article 113 AI Act and the May 2026 provisional AI Omnibus agreement (pending adoption):

Provision	Original AI Act date	AI Omnibus revised date (provisional)
Article 6(2) + Annex III standalone	2 August 2026	2 December 2027
Article 6(1) + Annex I product-safety	2 August 2027	2 August 2028
Annex III non-high-risk registration	(would have been removed under Commission Omnibus proposal)	Reinstated per provisional agreement

Article 111(2) AI Act provides that pre-2 August 2026 high-risk AI systems are not subject to the AI Act's high-risk requirements unless they undergo significant changes in design after that date; providers / deployers for public-authority use must meet the requirements by 2 August 2030 in any event. Article 111(1) provides a special regime for large-scale-IT components under Annex X (compliance by 31 December 2030).

The Omnibus deal is not adopted law as of the date of this page — see the EU AI Act overview for the conditional-outcome detail.

Settled vs vague (per draft Commission guidance)

Settled in the draft (subject to consultation):

• The four classification routes (A/B/C/D) and their relationships (paragraph (7) general principles).
• The four Article 6(3) conditions are alternative, not cumulative (paragraph (88) Annex III annex).
• The profiling override is absolute — Annex III systems performing profiling are always high-risk (third sub-paragraph of Article 6(3); paragraph (89)).
• Article 6(3) applies only to Article 6(2) Annex III systems, never to Article 6(1) Annex I systems (paragraph (87)).
• Intended purpose is the controlling concept; broad multi-purpose framing will be treated as encompassing high-risk uses (paragraph (12)).
• Article 6(4) documentation + Article 49(2) registration apply to non-high-risk Annex III determinations.

Left vague:

• The boundary of 'narrow procedural task' in Article 6(3)(a) is the central judgement call. The draft provides examples but no quantitative threshold; consultation feedback may sharpen this.
• The 'preparatory task' condition in Article 6(3)(d) overlaps with the others and is not separately defined in detail.
• The interplay between Article 6(1) and Article 6(2) when the same system meets both routes (e.g. an Annex III emotion-recognition system embedded in a medical device under Annex I Section A) — the draft treats these as separate triggers with parallel obligations.
• Substantial modification thresholds under Article 25(1)(b) are left to the forthcoming Article 25 guidelines.
• Sectoral overlap resolution under the AI Omnibus implementing acts (machinery, medical devices) is not yet defined.

How to operationalise high-risk classification in Modulos

Modulos models the Article 6 classification decision through two requirements on the MFF-1 framework template:

Article	Commission guidance section	Modulos requirement	Code
Article 3(1) AI-system test (prerequisite)	Definition guidance C(2025) 5053 (final)	AI System Classification (app)	MRF-38
Article 6(1) + Annex I product-safety route	Annex I annex sections III–IV	AI System Classification (app)	MRF-38
Article 6(2) + Annex III standalone route	Annex III annex sections III + 3.1–3.8	AI System Classification (app)	MRF-38
Article 6(3) filter mechanism	Annex III annex section 2.7 + paragraph (89) profiling override	AI System Classification Exemption (app)	MRF-111
Article 6(4) documentation	Annex III annex section 2.7.3, paragraphs (113)–(116) + Article 6(4)	AI System Classification Exemption (app) — documentation evidence	MRF-111
Article 49(2) EU-database registration	Article 49(2) cross-reference; AI Omnibus reinstated this	AI System Classification Exemption (app) — registration evidence	MRF-111
Article 25(1)(b)/(c) substantial modification	Paragraph (14) general principles + separate Article 25 guidelines (pending)	AI System Classification (app) — re-assessment	MRF-38

The classification rationale (which check applies, why, the intended-purpose statement, the Article 6(3) filter self-assessment if invoked, and the profiling determination) is owner-authored documentation stored as evidence on MRF-38. Where the Article 6(3) filter mechanism is invoked, the documentation + EU-database registration evidence sits on MRF-111.

For full template details see Operationalizing the EU AI Act in Modulos.

Cross-framework mapping (preview)

Article 6 element	Adjacent reading
Article 3(1) AI-system prerequisite	OECD updated AI-system definition (2024); ISO/IEC 22989.
Article 6(1) Annex I Section A NLF route	Sectoral conformity-assessment regimes (MDR/IVDR, Machinery, etc.); Blue Guide on implementation of EU product rules.
Article 6(2) Annex III use cases	NIST AI RMF MAP function (context-of-use characterisation); national risk-based AI laws (Colorado SB 24-205, Canada AIDA draft, Brazil PL 2338/2023).
Article 6(3) filter mechanism	OECD AI Principles; ISO/IEC 23894 risk management for AI.
Profiling override (Article 4(4) GDPR cross-reference)	GDPR Article 22 (right against decisions based solely on automated processing); EDPB Guidelines on automated individual decision-making.
Article 25 substantial modification	NIS2 / DORA substantial-change triggers (different thresholds, parallel logic).

Related pages

Commission guidance overview

Hub: legal status of Commission guidance, the five source documents, how to use them

High-risk worked examples (Commission draft)

Annex I sectoral examples + Annex III per-sub-point examples — draft guidance

Definition of an AI system (Commission guidance)

The Article 3(1) entry test — the prerequisite for any Article 6 analysis

Prohibited practices (Commission guidance)

For systems that meet Article 5 prohibitions, Article 6 classification does not apply

High-risk AI systems

The underlying Regulation text on Article 6 and the Articles 8–15 obligations

Operationalizing in Modulos

OFF-1 + MFF-1; the scoping questionnaire that records the Article 6 analysis

Source attribution

Draft Commission Guidelines on the classification of high-risk AI systems under Article 6 of the EU AI Act for stakeholder consultation, May 2026 — three companion documents: general principles, Article 6(1) Annex I annex, Article 6(2) Annex III annex. © European Union. Draft guidelines issued pursuant to Article 6(5) AI Act; not yet adopted. Section references on this page (e.g. "section 2.7") are to the numbered sections of the three companion documents; paragraph references in the form (N) are to the Commission's paragraph numbering. Draft — consultation closes 23 June 2026; text may change before formal adoption. The underlying Regulation (EU) 2024/1689 is published in the OJ L of 12 July 2024 under CELEX 32024R1689.

Disclaimer

This page is for general informational purposes and does not constitute legal advice.