Is AI exam proctoring high-risk under the EU AI Act?

Usually yes. If an AI system watches students during a test at a school, university or vocational training institution and flags behaviour it treats as cheating, it fits the wording of Annex III, point 3(d) almost word for word. Under Article 6(2) that makes it high-risk, unless the Article 6(3) derogation applies, and a system that profiles students cannot use the derogation. The answer turns on three things: where the test happens (the Annex III point only covers "educational and vocational training institutions", a term the Act does not define), what the system infers (inferring emotions in education is not high-risk but prohibited), and what the output is used for (in a hiring test, a different Annex III point may apply).

This is an information service to help you orient, not legal advice. To see the obligations that attach to your own AI use, answer three questions about it.

Your situation at a glance

Your situation	High-risk?	Why (article / annex point)
AI flags suspected cheating during exams at a school, university or vocational training institution	Yes	Annex III, point 3(d), read with Article 6(2)
The same system, but a human reviews every flag before anything happens	Usually still yes	Point 3(d) describes the system's intended purpose, not who decides. Article 6(3) needs a listed condition, and it never applies where the system profiles natural persons
Proctoring for a private certification or professional exam	Unclear	Point 3(d) applies only within "educational and vocational training institutions", a term the Act does not define
Proctored online test used to screen job applicants	Probably yes, by a different route	Not point 3(d), but if the flag feeds the evaluation of candidates, Annex III, point 4(a) can apply
A company's internal training course with an AI-proctored quiz	Unclear	Point 3(d) only if the company is a vocational training institution; point 4(b) may apply if results affect people's work
Webcam AI infers students' emotions (stress, nervousness) during a test	Prohibited	Article 5(1)(f), in force since 2 February 2025
The same emotion inference in a hiring or workplace test	Prohibited	Article 5(1)(f) also covers the workplace
Gaze or head-position tracking only, no emotion inferred	Yes, as proctoring	Point 3(d). The Commission's guidelines say gaze tracking to detect unauthorised material is not prohibited. Inferring an intention to cheat raises a separate open question (below)
One-to-one face match to confirm the candidate's identity	Not under point 1(a)	Point 1(a) excludes verification whose sole purpose is confirming a claimed identity. A proctoring system that includes it is still assessed under point 3(d)
AI that infers test-takers' race, religion or sexual orientation from their face	Prohibited	Article 5(1)(g), not limited to education or the workplace

The legal test

Article 6(2) says that AI systems referred to in Annex III "shall be considered to be high-risk". Annex III, point 3 covers education and vocational training. Point 3(d) lists:

"AI systems intended to be used for monitoring and detecting prohibited behaviour of students during tests in the context of or within educational and vocational training institutions at all levels."

That is a description of AI exam proctoring. Three elements do the work:

• "Intended to be used" points to the purpose the provider gives the system. A general video tool is not caught because someone could use it in an exam; a tool sold for exam integrity is.
• "Prohibited behaviour of students during tests" covers cheating detection. Grading is a separate listing, point 3(b), so a system that proctors and marks can fall under both.
• "Educational and vocational training institutions at all levels" covers schools and universities. The Act does not define the phrase, so a certification body, test centre or company academy has to argue from its ordinary meaning. Treat that question as open.

Recital 56 gives the reason: these systems "may determine the educational and professional course of a person's life". The Commission's Article 6 guidelines, required by Article 6(5), exist only as a draft published for consultation on 19 May 2026.

The Article 6(3) derogation, and why profiling closes it

Article 6(3) lets an Annex III system escape the high-risk label if it "does not pose a significant risk of harm" to health, safety or fundamental rights. It applies where at least one of four conditions is met. The system is intended to:

• (a) perform a narrow procedural task;
• (b) improve the result of a previously completed human activity;
• (c) detect decision-making patterns or deviations from prior patterns, without replacing or influencing the previously completed human assessment without proper human review; or
• (d) perform a preparatory task to an assessment relevant to an Annex III use case.

A tool that only flags students for a human to review can look like a preparatory task under (d). Two things make that a hard argument.

First, profiling. Under the third subparagraph of Article 6(3), an Annex III system "shall always be considered to be high-risk where the AI system performs profiling of natural persons". Article 3(52) borrows the GDPR definition, which covers automated processing of personal data to evaluate personal aspects of a person, in particular to analyse or predict their reliability or behaviour (Article 4(4) GDPR). A system that judges whether one student's behaviour looks like cheating is very likely doing that.

Second, the detection is the use case. Point 3(d) lists "monitoring and detecting", so the system doing the detecting is hard to describe as preparing for some other assessment.

If you still conclude a system is not high-risk, Article 6(4) requires you to document the assessment before placing it on the market or putting it into service, and Article 49(2) requires registration in the EU database. See when an Annex III system is not high-risk.

Article 5 prohibitions that can bite

Some proctoring features are not high-risk, because they are banned. The Article 5 prohibitions have applied since 2 February 2025. Under Article 99(3), breaking them carries fines of up to EUR 35 million or, for an undertaking, 7% of total worldwide annual turnover for the preceding financial year, whichever is higher.

Emotion inference — Article 5(1)(f). The Act prohibits the placing on the market, putting into service for this specific purpose, or use of AI systems "to infer emotions of a natural person in the areas of workplace and education institutions", except where the system is intended for medical or safety reasons. A proctoring feature that reads stress, anxiety or nervousness from a student's face or voice is squarely in the sights of this ban. The Commission's guidelines on prohibited practices (C(2025) 5052 final, paragraph 255), which are not binding, give two education examples: an institution using eye tracking during online exams to detect whether unauthorised material is used is not prohibited, because the system infers no emotion, but it would be if the system were also used to detect emotions such as emotional arousal or anxiousness; and using emotion recognition to infer students' interest and attention is prohibited. For where the line falls, see emotion recognition at work and in education.

An open question: inferring intentions. Article 3(39) defines an emotion recognition system as one that identifies or infers "emotions or intentions" from biometric data, and Article 3(34) counts behavioural characteristics as biometric data. A model that infers an intention to cheat from eye or head movement may therefore be an emotion recognition system. That would bring in Annex III, point 1(c), and the Article 50(3) duty to inform. Article 5(1)(f), however, says only "infer emotions". The Commission's guidelines (paragraph 245) read the prohibition as covering AI systems that identify or infer emotions or intentions, but the guidelines are not binding and only the Court of Justice can settle the point. Whether intention inference in education is prohibited, or "only" high-risk, is therefore not settled, though on the Commission's reading it leans towards prohibited. Design and document with both readings in mind.

Biometric categorisation — Article 5(1)(g). This bans systems that categorise individual people from their biometric data to deduce or infer their race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation. Proctoring has no reason to do this; if a model does it as a by-product, the ban is not limited to education or the workplace.

For the full list, see EU AI Act prohibited practices (Article 5).

If it is high-risk: who does what, and when

When. The Digital Omnibus on AI (Regulation (EU) 2026/1744) amended Article 113. Chapter III, Sections 1, 2 and 3 now apply to Annex III systems from 2 December 2027.

Providers (who develop the system and place it on the market under their own name) carry the main load under Article 16: the Section 2 requirements (risk management, data governance, technical documentation, logging, human oversight, accuracy, robustness and cybersecurity), a quality management system, a conformity assessment (internal control under Annex VI for point 3 systems, Article 43(2)), an EU declaration of conformity, CE marking and registration in the EU database (Article 49(1)).

Deployers (the school, university, certification body or employer using it) have their own duties under Article 26: use it according to the instructions for use, assign human oversight to people with the necessary competence, training and authority, monitor it, keep the logs under their control for at least six months, and tell the people affected that a high-risk system is used on them (Article 26(11)). Employers must also inform workers' representatives and affected workers first (Article 26(7)). See deployer obligations under Article 26.

Public institutions. Deployers that are public authorities must register their use in the EU database (Article 49(3)) and must not use an unregistered system (Article 26(8)). Under Article 27, bodies governed by public law and private entities providing public services must carry out a fundamental rights impact assessment before first use. Whether a private university "provides public services" depends on its circumstances.

Becoming a provider by accident. Under Article 25(1), a deployer that puts its own name on a high-risk system, substantially modifies one, or turns a general tool into a proctoring system takes on the provider's obligations.

What applies regardless

• AI literacy (Article 4). The AI literacy duty has applied since 2 February 2025. As amended by the Digital Omnibus, it requires providers and deployers to take measures to support the AI literacy of staff dealing with the system, such as the people who review flags, and it does not require them to guarantee any specific level of literacy. See the AI literacy duty.
• Transparency (Article 50(3)). Since 2 August 2026, deployers of an emotion recognition or biometric categorisation system must inform the people exposed to it. That can matter for intention inference or outside education. See transparency obligations.
• Data protection. Webcam proctoring processes personal data, often biometric data. The AI Act does not replace the GDPR.

What to do now

1. List every proctoring feature separately: identity check, gaze tracking, audio monitoring, behaviour flags, any "stress", "engagement" or "suspicion" score.
2. Remove or switch off emotion inference in any education or workplace setting. The ban is already in force.
3. Decide your setting honestly. Are you an educational or vocational training institution? If not, does the result feed a hiring or employment decision?
4. Assume high-risk for cheating detection and plan for 2 December 2027. If you claim Article 6(3), check first whether the system profiles anyone, and write your reasoning down.
5. Public bodies: plan the fundamental rights impact assessment and the database registration now.

For the general test behind this page, see is your AI system high-risk?. To get the obligations that match your own use, sorted by deadline, start the obligation check. To get a weekly email of verified changes to these rules, join the waitlist.

The official text is Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744. This article is an information service to help you orient — it is not legal advice.