Is your AI system high-risk? A decision tree

Updated 9 August 2026 for Regulation (EU) 2026/1744.

Answer five yes-or-no questions in order and you will know whether your AI system is high-risk under the EU AI Act. Have the system's intended purpose in front of you before you start. The five are the Article 5 prohibitions, the Article 6(1) product route, Annex III under Article 6(2), the Article 6(3) exemption, then Article 50 transparency. Work down the tree; the first answer that sticks is your tier among the four outcomes.

Run it with the system's intended purpose to hand; the full method — and each tier's paperwork — is the classification guide.

Question zero is not on the tree: every branch below presumes the thing being classified meets the Article 3(1) definition of an AI system. If a rules engine or a hand-tuned scoring formula has you unsure whether you are even holding one, settle that first — the tiers grade AI systems, nothing else.

The tree

1. Does the system do anything Article 5 prohibits? Ten practices now. Eight — subliminal or manipulative techniques, exploiting vulnerability, social scoring among them — have been banned since 2 February 2025 (Article 113, third paragraph, point (a)). Points (ba) and (bb), on non-consensual intimate material and child sexual abuse material, were inserted by Regulation (EU) 2026/1744 and apply from 2 December 2026 (Article 113, third paragraph, point (a), as amended). Read them with the new Article 5(1a), which limits the placing-on-the-market and putting-into-service ban for both points to systems whose intended purpose is that generation or manipulation or where it is a reasonably foreseeable and reproducible outcome without adequate safeguards, and limits the use ban to a deployer who uses the system for that purpose; and with Article 5(1b), which for point (ba) alone excludes manipulation that does not increase the exposure of any depicted intimate parts or alter the nature of any depicted sexually explicit activities.

• YES → Stop — prohibited is not a risk tier; no exemption or registration makes it lawful. Check your match in which AI practices are prohibited.
• NO → Question 2.

2. Is it a product, or the safety component of one, under Article 6(1)? Both conditions must hold: covered by the Union harmonisation legislation listed in Annex I, and required to undergo third-party conformity assessment under it. Regulation (EU) 2026/1744 narrowed both. Article 6(1a) puts outside "safety component" systems "solely used for non-safety related aspects of user assistance, performance optimisation, service efficiency, automation or convenience or quality control"; Article 6(1b) then makes systems "the failure or malfunctioning of which would endanger health and safety" qualify notwithstanding 6(1a). Article 6(1c) fails the second condition where third-party assessment is required "solely due to risks other than risks to health and safety", naming radio-spectrum and electromagnetic-interference risks that do not affect health and safety. The same amendment moved the Machinery legislation from point 1 of Section A of Annex I (Directive 2006/42/EC, deleted) to point 21 of Section B (Regulation (EU) 2023/1230).

• YES, both → High-risk under Article 6(1), applying from 2 August 2028 (Article 113, third paragraph, point (c), as amended by Regulation (EU) 2026/1744). Continue with the classification method.
• NO, either limb fails → Question 3.

3. Does the intended use match a specific point of Annex III (Article 6(2))? Match at point-and-letter level — "we touch employment data" is not a match; "we filter job applications, point 4(a) of Annex III" is. Read the point's carve-outs too — credit scoring carries one, for financial-fraud detection.

• YES → Question 4.
• NO → Question 5.

4. Does the Article 6(3) exemption take it back out? In this order:

• The system performs profiling of natural persons → High-risk, always — profiling voids the exemption. Locate your point in the Annex III reference.
• No profiling, and one of the four Article 6(3) conditions holds — a narrow procedural task; improving the result of a previously completed human activity; detecting decision-making patterns or deviations from them, not meant to replace or influence the previously completed human assessment, without proper human review; a preparatory task to an assessment relevant to the Annex III use cases → Possibly exempt — but only with the Article 6(4) assessment documented before market placement and Article 49(2) registration. The classification method owns that test.
• Neither → High-risk under Article 6(2), obligations from 2 December 2027 (Article 113, third paragraph, point (c), as amended by Regulation (EU) 2026/1744). Find your exact point and letter in the Annex III reference.

5. Does Article 50 transparency attach? It does if the system interacts directly with natural persons (Article 50(1)), generates synthetic audio, image, video or text (Article 50(2)), performs emotion recognition or biometric categorisation (Article 50(3)), or produces deep fakes (Article 50(4)).

• YES → Limited risk: disclosure and marking duties — see the limited-risk transparency guide. They layer onto high-risk too (Article 50(6)); a "yes" here cancels nothing earlier.
• NO → Minimal risk. No tier-specific obligations attach — the Article 4 AI-literacy duty applies at every tier; record the reasoning anyway. What minimal risk actually leaves covers both.

What this means for you

If you're a provider: The tree is a first answer, not a classification — the call you stand behind is documented per system, above all any Article 6(4) exemption assessment. An Annex III high-risk answer means obligations from 2 December 2027 — see what the high-risk deadline triggers. Complipath's guided risk classification returns a risk level with the provisions and reasoning it rests on.

If you're a deployer: Run the tree against your own intended use — it can differ from the provider's. If a vendor claims the question 4 exemption, ask for the Article 6(4) assessment — it must predate market placement.

Get the answer for your own system

Classify your system now — 7 questions on the main line, plus follow-ups where they apply, no account, and the classification runs in your browser: answers stay there unless you choose to keep the result.

FAQ

Is the decision tree enough to classify my system? No — it gives the first answer, fast. A defensible classification is documented per system at every tier; claiming the Annex III exemption requires the Article 6(4) assessment before market placement plus Article 49(2) registration. The full method is in the classification guide.

Does human review take my system out of high-risk? Not by itself. Article 6(3) exempts only through its four conditions, and condition (c) covers only systems not meant to replace or influence the completed human assessment without proper human review. No condition survives profiling: an Annex III system that profiles natural persons is always high-risk.

When does each answer start to bite? Eight Article 5 prohibitions have applied since 2 February 2025; points (ba) and (bb) apply from 2 December 2026. Annex III high-risk obligations apply from 2 December 2027, Article 6(1) product classifications from 2 August 2028 (Article 113, third paragraph, points (a) and (c), as amended by Regulation (EU) 2026/1744).

Can one system get two answers from the tree? Yes — Article 50 is a layer, not a tier. Its duties attach to specific behaviours whatever the classification, and Article 50(6) provides that they do not affect the Chapter III requirements. A high-risk system with a conversational interface carries both sets.

Sources: Regulation (EU) 2024/1689 (EUR-Lex), Articles 4, 5, 6, 49, 50 and 113, and Annexes I and III, as amended by Regulation (EU) 2026/1744 (EUR-Lex) — in force 27 July 2026 — which inserted Article 5(1), first subparagraph, points (ba) and (bb), Article 5(1a) and (1b) and Article 6(1a), (1b) and (1c), replaced Article 113, third paragraph, points (a) and (c), and moved the Machinery legislation from point 1 of Section A of Annex I (Directive 2006/42/EC, deleted) to point 21 of Section B (Regulation (EU) 2023/1230). That move has a consequence this tree does not yet work through: under Article 2(2), as also replaced by Regulation (EU) 2026/1744, "for AI systems classified as high-risk AI systems in accordance with Article 6(1) related to products covered by the Union harmonisation legislation listed in Section B of Annex I, only Article 6(1), Article 60a and Articles 102 to 112 shall apply" — so a YES at question 2 for a machinery product is still high-risk but carries a reduced obligation set. Article 4 was also replaced by that Regulation; the AI-literacy duty still applies with no risk-level qualifier, but its wording is now "support the development of AI literacy" rather than "ensure, to their best extent, a sufficient level".