Worked example: a classroom proctoring tool

What are the invented facts?

The invented facts are stated here so that the screening can be tested against them. A school, Deployer P, uses a proctoring tool, Tool Q, supplied by a vendor, Provider R, to monitor students during an online exam. Tool Q uses a webcam and a microphone to record the student, to detect eye movement, head movement and background voices, to flag behaviour that may indicate cheating, and to produce a report for a teacher. The teacher reviews the report, watches the recording where a flag is raised, and decides whether any action is taken, including whether to require a retake or to start a disciplinary process. Tool Q does not infer emotions, on the invented facts, and does not identify students by biometric identification. It verifies identity at the start of the exam by comparing a photo on file with the webcam image, on the invented facts, which raises a separate biometric question that must be examined. The tool is placed on the Union market by Provider R and put into service by Deployer P in the Union, and is used for exams in the Union. The students are minors in some cases, on the invented facts, which raises additional data protection, child protection and fundamental rights questions. Those facts are invented and simple, deliberately, so that the screening steps are visible. A real proctoring tool will have more complex facts, including the data retention, the accuracy across different students, the accessibility, the human review design, and the appeal arrangements, each of which can change the screening.

How is scope and role screened?

The facts that matter most here are the ones about seeing and hearing students. Tool Q records webcam and microphone input, detects eye movement, head movement and background voices, flags behaviour that may indicate cheating, and produces a report that a teacher reviews before deciding whether to require a retake or start a disciplinary process. Identity is checked at the start of the exam by comparing a photo on file with the webcam image, which raises its own biometric question. Those facts give the roles directly: Provider R develops the tool and places it on the market under its own name, and Deployer P, the school, uses it under its own authority for an educational activity, with the school also likely to be a public body or a body providing a public service, which matters for the Article 27 assessment and for data protection and education law. Tool Q meets the Article 3(1) definition on these facts, subject to the Commission guidelines, and is placed on the Union market and used in the Union, so Article 2 is satisfied. Deployer P takes on no provider duties on these facts, but that must be rechecked if the school customises the tool or changes its purpose.

For the broader regulatory framework, see questions and misconceptions overview.

How is the prohibition screen done?

Two prohibitions need the closest attention, and both turn on what the tool infers rather than what it is called. Article 5 prohibits AI systems that infer emotions of a natural person in education, with defined exceptions, so the question is whether detecting eye movement, head movement and background voices amounts to inferring stress, anxiety, attention or engagement from facial, voice or behavioural data, even where the feature is described as attention detection or focus monitoring. The Article 5 text, the Article 3 definitions and the Commission guidelines on prohibited practices must be read together on that point, and a tool that crosses the line must not be used. Biometric identification needs the same element-by-element test: comparing a photo with a webcam image must be examined against the biometric identification and categorisation prohibitions, including which data are used, for what purpose, and with what safeguards. The new December 2026 prohibitions are not shown on these facts. Any conclusion is provisional, tied to the invented facts, and needs qualified advice before use.

How is the Annex route screened?

Annex III, point 3 covers AI systems intended to determine access or admission, assign persons to institutions, evaluate learning outcomes, assess the appropriate level of education, and monitor and detect prohibited behaviour of students during tests. Tool Q monitors and detects prohibited behaviour during tests and evaluates learning outcomes through the exam, so it falls within that use case in outline. Article 6(3) is then tested and fails on these facts: the tool does more than a narrow procedural task, more than improving a prior human result, more than detecting decision-making patterns without replacing human assessment, and more than a preparatory task, because it flags behaviour, produces a report and shapes the teacher's decision. Profiling closes the route in any event, since the tool profiles students by flagging and reporting their behaviour. The provisional conclusion, tied to the invented facts, is that Tool Q is a high-risk candidate under Article 6(2), with the classification record, the prohibition analysis and the Article 27 position to be completed and confirmed before any use.

What rights, data and accessibility questions remain open?

The open questions for Tool Q include the data protection impact assessment, the lawful basis and the consent position for students, including minors, the data retention and access arrangements, the accuracy and bias testing across different students, including students with disabilities, different skin tones, different languages and different home environments, the accessibility of the tool, the human review design and the teacher training, the appeal and redress arrangements for a student who disputes a flag, the information to students and parents, the Article 27 fundamental rights impact assessment, the post market monitoring plan, and the application date for the high risk duties, which for Annex III systems is 2 December 2027. Each open question needs an owner, a date and evidence. The example also leaves open whether the identity verification function is a separate system or a function of Tool Q, whether the webcam and microphone recording is proportionate, and whether a less intrusive alternative, such as an in person exam or a different assessment design, should be used. A proctoring file that does not show the rights, data and accessibility evidence, and the consideration of less intrusive alternatives, will be difficult to defend, even where the classification is clear. The screening record should list the open questions with the action taken, and should be updated as the evidence is produced.

Continue with prohibited practices in Article 5.

What does this example not decide?

Nothing here settles a real proctoring tool. The emotion inference and biometric positions for a named tool, and any data protection, education law or child protection opinion for a school, are outside this walkthrough, which is hypothetical and illustrative, built from invented facts, with a provisional conclusion tied to those facts and a confirm-with-authority exit. A provider or deployer with a real proctoring system should run the screening order on its own facts, check the consolidated Regulation and the Commission guidelines, and confirm the prohibition and classification positions with a competent authority or legal adviser before any use. The method is transferable; the answer for a real system is not.

Frequently asked questions

Continue through this AI Act cluster

• questions and misconceptions overview
• prohibited practices in Article 5
• screening order and evidence record
• healthcare and biometrics
• high risk requirements
• worked example: recruitment screening