What is the difference between Annex I and Annex III of the EU AI Act?

Annex I lists EU product legislation, Annex III lists use cases. Under Article 6(1), an AI system intended to be used as a safety component of a product under an Annex I act, or that is itself such a product, is high-risk where the product needs third-party conformity assessment; under Article 6(2), an Annex III system is, unless Article 6(3) exempts it. Chapter III, Sections 1, 2 and 3, except Article 6(5), apply on the first route from 2 August 2028 and on the second from 2 December 2027 (Article 113, third paragraph, point (c), as replaced by Regulation (EU) 2026/1744).

Written and last checked 9 October 2026 against Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744. Article 6(1) to (4) and Annex III were not amended; Annex I was.

When is a system high-risk under Annex I?

When both conditions of Article 6(1) hold. Point (a): "the AI system is intended to be used as a safety component of a product, or the AI system is itself a product, covered by the Union harmonisation legislation listed in Annex I". Point (b): that product "is required to undergo a third-party conformity assessment", with a view to its placing on the market or putting into service under that legislation. The chapeau applies this "Irrespective of whether an AI system is placed on the market or put into service independently of the products referred to in points (a) and (b)".

Regulation (EU) 2026/1744 qualified both conditions without amending Article 6(1). Article 6(1a), inserted, says 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) brings back "AI systems the failure or malfunctioning of which would endanger health and safety".

Article 6(1c) says a product that must undergo third-party conformity assessment "solely due to risks other than risks to health and safety" does not fulfil point (b). Article 3, point (14), as amended, adds that a component fulfils a safety function "where its intended purpose is to prevent or mitigate risks to health and safety of persons or property".

When is a system high-risk under Annex III?

Article 6(2) is one sentence: "In addition to the high-risk AI systems referred to in paragraph 1, AI systems referred to in Annex III shall be considered to be high-risk." Annex III lists eight areas, all but point 2 split into lettered use cases; a system is caught when its intended purpose matches one. Neither an Annex I product nor a third-party conformity assessment enters the test. The Annex III guide covers each point.

Does the Article 6(3) exemption work for both?

No. Article 6(3) opens: "By derogation from paragraph 2, an AI system referred to in Annex III shall not be considered to be high-risk where it does not pose a significant risk of harm to the health, safety or fundamental rights of natural persons, including by not materially influencing the outcome of decision making." It needs one of the four conditions in its second subparagraph, and an Annex III system that performs profiling of natural persons is "always" high-risk.

The Article 6(3) guide covers the conditions and what a provider relying on them owes under Article 6(4). Article 6(1) has no derogation of its own: an Annex I system is on the route when both conditions hold, read with Article 6(1a), (1b) and (1c).

When does each route apply?

Article 113, third paragraph, point (c), as replaced, reads: "Chapter III, Sections 1, 2, and 3, with the exception of Article 6(5), shall apply from: (i) 2 December 2027 as regards AI systems classified as high-risk pursuant to Article 6(2) and Annex III; and (ii) 2 August 2028 as regards AI systems classified as high-risk pursuant to Article 6(1) and Annex I". The point it replaced applied "Article 6(1) and the corresponding obligations in this Regulation" from 2 August 2027, and left Annex III systems on the general date of 2 August 2026.

Point (c) omits Sections 4 and 5. Section 4 has applied since 2 August 2025 (Article 113, third paragraph, point (b)); Section 5, which holds conformity assessment and registration, takes the general date of 2 August 2026 (Article 113, second paragraph). Neither text says how it operates before Section 1 applies. The deadlines table carries every date.

What if a system is on both routes?

Article 43(3), as replaced, answers the conformity assessment question: "Where a high-risk AI system is both covered by the Union harmonisation legislation listed in Section A of Annex I and it falls within one of the categories listed in Annex III, the provider of that system shall follow the relevant conformity assessment procedure as required pursuant to the relevant Union harmonisation legislation listed in Section A of Annex I." The 2024 text had no such subparagraph. Neither text says which date in point (c) then governs; this guide does not resolve it.

What else differs between the routes?

• Conformity assessment: Annex III points 2 to 8 follow internal control under Annex VI, with no notified body (Article 43(2)), and point 1 follows Article 43(1). Products under Section A of Annex I follow their own legislation's procedure, with the Section 2 requirements part of that assessment (Article 43(3), as replaced); the conformity assessment guide has the detail.
• Registration: Article 49(1) and (3) cover high-risk systems listed in Annex III, except point 2, which is registered at national level (Article 49(5)). Article 49 names no registration for the Annex I route.
• Deployers: the fundamental rights impact assessment covers systems "referred to in Article 6(2)", outside point 2 of Annex III and only for the deployers it names (Article 27(1)). Under Article 26(11), deployers of systems "referred to in Annex III" that make or assist in making decisions related to natural persons inform them that they are subject to its use; for law enforcement, Article 13 of Directive (EU) 2016/680 applies.
• Limited requirements: for high-risk systems referred to in Article 6(1), the requirements and obligations of Articles 9 to 15 and 17 to 25 may be limited where legislation in Section A of Annex I gives an equivalent or higher level of protection and the overall level of protection is not reduced (Article 2(13), as added). Article 2(13) has the Commission specify this by delegated acts by 2 August 2027, and none is in our pinned sources.

What changed in Annex I in 2026?

Regulation (EU) 2026/1744, Article 1, point (41), made two changes: "in Section A, point 1 is deleted", which was Directive 2006/42/EC on machinery; and Section B gains point 21, Regulation (EU) 2023/1230 on machinery. Points 2 to 20 were not amended.

This matters under Article 2(2), as replaced: for AI systems classified as high-risk under Article 6(1) and related to products under Section B, "only Article 6(1), Article 60a and Articles 102 to 112 shall apply", and Articles 57, 58 and 59 only in so far as the high-risk requirements have been integrated in that legislation. The 2024 text it replaced read "only Article 6(1), Articles 102 to 109 and Article 112" and named Article 57 alone.

For machinery, Article 3 of Regulation (EU) 2026/1744 has the Commission add health and safety requirements for these AI systems to Regulation (EU) 2023/1230 by delegated acts, which "shall apply by 2 August 2028". This guide has not checked whether any has been adopted.

What does Complipath do here, and what does it not?

The rule engine does not detect Article 6(1): its assessment has no Annex I question. The Annex I high-risk check, in Starter and the trial, asks the three facts the route turns on, each beside the limb of Article 6(1) it rests on, and takes Yes, No or I don't know. The record carries who answered and when, and that Complipath has not verified the answers; an I don't know leaves the route undecided, with what you would need to find out.

It neither changes the classification nor derives the requirements that follow from Annex I. Risk classification covers the Annex III side and the full list of features the rest; AI register tools ranked compares other vendors by their own pages.

What this means for you

If you're a provider: run both tests on every system. For a product, name the act in Annex I and its section, and check point (b) with Article 6(1c) beside it. For a use case, name the point of Annex III and, where it has one, the letter; only there can Article 6(3) apply.

If you're a deployer: ask the provider which route each system is on; it decides which of your duties apply and from when. Article 27, Article 26(11) and Article 49(3) attach to Annex III systems only.

FAQ

Can an AI system be high-risk under both Annex I and Annex III? Yes. Article 6(2) applies "in addition to" the systems referred to in paragraph 1. Where a system is covered by legislation in Section A of Annex I and falls within an Annex III category, Article 43(3), as replaced, sends the provider to that legislation's conformity assessment procedure. Which application date governs it, the text does not say.

Does the Article 6(3) exemption apply to Annex I systems? No. Article 6(3) is a derogation from paragraph 2, the Annex III route. It covers only "an AI system referred to in Annex III". A system on the Annex I route stays high-risk while both conditions of Article 6(1) hold. It can only fall outside the route by failing one of them, read with Article 6(1a), (1b) and (1c).

When do the obligations apply for each route? Chapter III, Sections 1, 2 and 3, except Article 6(5), apply from 2 December 2027 for systems high-risk under Article 6(2) and Annex III, and from 2 August 2028 under Article 6(1) and Annex I (Article 113, third paragraph, point (c), as replaced). Sections 4 and 5 keep their own dates, 2 August 2025 and 2 August 2026.

Does every AI feature in a product under Annex I count as a safety component? No. Article 6(1a) excludes systems solely used for non-safety related aspects of user assistance, performance optimisation, service efficiency, automation or convenience or quality control. Article 6(1b) brings back any system whose failure or malfunctioning would endanger health and safety. A product needing third-party assessment solely for risks other than to health and safety fails point (b) (Article 6(1c)).

Sources: Regulation (EU) 2024/1689 (EUR-Lex), Article 2(2), Article 3, point (14), Article 6, Article 26(11), Article 27(1), Article 43, Article 49, Article 113 and Annexes I and III; as amended by Regulation (EU) 2026/1744 (EUR-Lex), which replaced Article 2(2), Article 43(3) and Article 113, third paragraph, point (c), added Article 2(13), amended Article 3, point (14), inserted Article 6(1a), (1b) and (1c) and amended Annex I (Article 1, points (2), (3), (4), (8), (19), (40) and (41)), and whose Article 3 amends Regulation (EU) 2023/1230. Article 6(1) to (4), Article 26, Article 27(1), Article 49 and Annex III were not amended. Whether the delegated acts under Regulation (EU) 2023/1230 have been adopted is not in our source corpus.

Where this question meets the product: the AI Act risk classification tool and what it records. Answering it in a customer's questionnaire: how to answer AI questions in a supplier questionnaire. The feature that does this in the app: every live Complipath feature. Tools ranked on it: Best AI register software for the EU AI Act.