Is your SaaS "high-risk" under the EU AI Act? Annex III, read calmly
If you sell HR tech, edtech, insurtech or credit-adjacent software with AI in it, someone — a consultant, a competitor's marketing, a worried buyer — has probably told you that you're "high-risk AI" and a mountain of obligations is coming. Sometimes that's true. More often, the claim is made without reading Annex III, which is a closed list of specific use cases, not a verdict on entire industries. Getting your classification right matters twice: once for the law, and once for the questionnaire where your buyer asks you to state it.
What Annex III actually lists (the areas relevant to SaaS)
Annex III attaches high-risk status to AI systems intended to be used for specific purposes within eight areas. The ones that touch commercial software vendors most often:
| Area | Listed use cases (paraphrased) | Typical vendor affected |
|---|---|---|
| Education (Annex III, 3) | Determining access or admission to education; evaluating learning outcomes; assessing the appropriate level of education; monitoring and detecting prohibited behaviour during tests. | Edtech doing admissions scoring, exam proctoring, high-stakes assessment. |
| Employment (Annex III, 4) | Recruitment and selection (placing ads, filtering applications, evaluating candidates); decisions on promotion, termination, task allocation based on behaviour or traits; monitoring and evaluating performance. | HR tech doing candidate screening/ranking, performance evaluation. |
| Essential services (Annex III, 5) | Creditworthiness evaluation and credit scoring of natural persons; risk assessment and pricing in life and health insurance for natural persons; among others. | Fintech credit scoring; insurtech pricing life/health risk of individuals. |
Note what is not there: AI that summarizes documents for recruiters, drafts job descriptions, answers employees' HR questions, recommends learning content, or automates insurance claims paperwork is not, by that fact alone, on the list. The question is always whether the system is intended for one of the listed purposes.
The carve-out vendors forget: Article 6(3)
Even within a listed area, a system is not considered high-risk if it does not pose a significant risk of harm to health, safety or fundamental rights — including where it is intended to perform a narrow procedural task, to improve the result of a previously completed human activity, to detect decision-making patterns without replacing or influencing the human assessment, or to perform a preparatory task to an assessment. A provider relying on this must document that assessment (and registration duties still apply). The honest reading: a tool that ranks candidates for a human recruiter's final decision is squarely in Annex III, 4; a tool that transcribes interviews is doing a preparatory/procedural task.
The dates, after the Digital Omnibus
- High-risk (Annex III) obligations were postponed to 2 December 2027 (embedded-in-regulated-products high-risk: 2 August 2028).
- The Article 50 transparency duties are separate and were NOT postponed — they apply from 2 August 2026 to chatbots, voice agents and synthetic-content features regardless of risk tier. Being outside Annex III does not exempt you from those. (See: Article 50, who owes what.)
Why your buyer asks before the regulator does
SIG 2026 and AI-CAIQ questionnaires, and enterprise RFPs, now ask vendors to state their EU AI Act risk classification. Answering "high-risk" when you're not signs you up for obligations you don't owe and alarms procurement; answering "not applicable" when you're plainly in Annex III, 4 tells the reviewer you haven't read the Regulation. Both cost deals. The classification answer is the one to get right first, because every other AI answer in the questionnaire hangs off it.
How to answer the classification question
- Name the use case, not the industry — "our system performs X", mapped (or not) to the specific Annex III entry.
- If in a listed area but relying on Article 6(3), say which limb (procedural / preparatory / pattern-detection) and note the assessment is documented.
- State the dates that bind you — Annex III duties from 2 December 2027 if applicable; Article 50 transparency from 2 August 2026 either way.
- Describe current measures and the roadmap — for genuinely high-risk systems, buyers accept "building toward the 2027 requirements, here's the plan" far better than silence.
- Regulation (EU) 2024/1689, Official Journal — Annex III; Art. 6(2)–6(3); Art. 50; Art. 113 (ELI: data.europa.eu/eli/reg/2024/1689/oj)
- Digital Omnibus: European Parliament position 16 June 2026; Council adoption 29 June 2026 (consilium.europa.eu press releases) — high-risk dates moved to 2 Dec 2027 / 2 Aug 2028; Art. 50 unchanged
- Shared Assessments — SIG 2026 (AI domain); Cloud Security Alliance — AI-CAIQ