By industry · Written by Yobel Tzegai · Last checked 8 November 2026

Is AI in education high-risk under the EU AI Act?

Education AI under the EU AI Act: admission, grading, level placement and proctoring are high-risk (point 3 of Annex III); emotion inference is banned.

Get my risk classes, freeBook a 20-min intro call20 minutes with the founder.

The answer, in detail

Where an education use usually sits
UseUsual outcomeProvisionApplies from
Admission, or assigning students to institutionsHigh-riskPoint 3(a) of Annex III2 December 2027, under Article 113, third paragraph, point (c), as replaced
Grading, or evaluating learning outcomesHigh-riskPoint 3(b) of Annex III2 December 2027, under the same point
Placing a student at a level of educationHigh-riskPoint 3(c) of Annex III2 December 2027, under the same point
Proctoring tests for prohibited behaviourHigh-riskPoint 3(d) of Annex III2 December 2027, under the same point
Inferring students’ or staff’s emotionsProhibited, except for medical or safety reasonsArticle 5(1), first subparagraph, point (f)2 February 2025, under Article 113, third paragraph, point (a), as replaced
A tutoring chatbotTransparency duties for its providerArticle 50(1); Article 50(2) for the text it generates2 August 2026, under Article 113, second paragraph; Article 50(2) by 2 December 2026 for a system placed on the market before 2 August 2026 (Article 111(4))

Which education systems are high-risk?

Who owes what, the school or the edtech company?

What applies to every use?

Your customer asks. You answer with proof.

  1. 1List every AI system

    Add the AI you use. Each system gets a role, a risk class and the article it rests on.

    Live
  2. 2Attach the proof

    Link each duty to a file and the page it stands on: a policy, a log setting, a training record.

    Live
  3. 3Answer their questionnaire

    Answering a customer's questionnaire inside the app, from the proof you already linked.

    Live

What Complipath does

  • Risk classification Answers go through rules in code, never a language model, so the same answers always give the same result. Rules decide. AI only drafts. A person confirms.
  • AI inventory Every AI system you build or use, with its owner and risk class.
  • AI literacy (Article 4) Records the measures you take to support AI literacy under Article 4 as amended: who was briefed on what, and when. It applies whatever your risk level.
  • Obligations per system Confirming a classification creates the requirements that follow from it, each with a status, a place for evidence and an owner you fill in: for high risk, the set for the system's role (provider, deployer or both, from its recorded type or confirmed role), and a row for each Article 50 transparency duty. The rows are made at a system's first confirmation; when a later high-risk classification, a role change or a newer catalogue leaves high-risk rows missing, Review proposes them and a person adds them. The limit, in the app's own words: "Rows appear once a person confirms the classification, and a person marks each one done, never the app".
  • Annex IV documentation A workspace for the technical documentation, section by section, saying which Annex IV points each one answers.
  • Fundamental rights impact assessment (Article 27) Two features. 'DPIA and FRIA: whether they apply' says for each confirmed system whether a fundamental rights impact assessment is required, likely or not, quoting Article 27(1). 'DPIA and FRIA documents', in the app's words, drafts the FRIA (Article 27(1)) from your register by rules, built on the DPIA, a source on every sentence, for a named person to confirm with a date: included in Growth.

Rules decide. AI only drafts. A person confirms.

What it does not do yet

  • Conformity assessment (Article 43) With limits Complipath does not carry out the conformity assessment. It drafts what the internal-control procedure in Annex VI checks: the Annex IV technical documentation part by part on every plan, and on Growth, for a provider of a high-risk system, the quality management system (Article 17) and the post-market monitoring plan (Article 72). On Growth it also drafts the EU declaration of conformity (Article 47, Annex V). A person completes and signs them.

What is live today

Checked against the app on 8 November 2026
Starter

€499 a month

or €4,990 a year. Prices exclude VAT.

  • Up to 20 AI systems
  • Evidence and the audit log
  • The Free plan is for one AI system and one user, with no end date
Get my risk classes, free

All plans

Questions

Is an AI tool that helps teachers grade essays high-risk?

If it is intended to evaluate learning outcomes at an educational institution, it is in point 3(b) of Annex III. Article 6(3) can take it out where, for example, it only improves the result of a previously completed human activity, but never where it profiles natural persons, and the provider documents that assessment before placing it on the market (Article 6(4)).

Can we use AI to detect whether students are paying attention?

Not by inferring their emotions. Article 5(1), first subparagraph, point (f) prohibits AI systems that infer the emotions of a natural person in education institutions, except where intended for medical or safety reasons. Monitoring for prohibited behaviour during tests is a different use, and high-risk under point 3(d) of Annex III.

Read next
See what your customers will ask about your AI.No account needed. Every answer cites the article it rests on.