AI Exam Proctoring Under the EU AI Act: High-Risk Classification and What Institutions Must Do Before December 2027

How the AI Act classifies AI proctoring

Regulation (EU) 2024/1689 states in its recitals that AI systems used in education or vocational training should be classified as high-risk. The listed uses include determining access or admission, assigning people to institutions or programmes, evaluating learning outcomes, assessing the appropriate level of education, and monitoring and detecting prohibited behaviour of students during tests. The stated rationale is that these systems may determine a person's educational and professional course of life and so affect their ability to secure a livelihood.

The Commission's FAQ adds the operative principle: risk classification is based on intended purpose. It depends on the function the system performs and on the specific purpose and modalities for which it is used. A proctoring tool whose intended purpose is to monitor and detect prohibited behaviour during tests matches the education use case directly. An AI system can be high-risk where it is intended for a use case listed in Annex III, which is the route relevant to proctoring software.

This means a vendor's description of its product does not settle classification. The intended purpose in the provider's documentation, and how the institution actually configures and uses the system, both matter.

The derogation and why the scoping record matters

The consolidated Act provides that an Annex III system is not considered high-risk where it does not pose a significant risk of harm to health, safety or fundamental rights, including by not materially influencing the outcome of decision making. Institutions considering this route should be cautious. A proctoring flag that feeds an academic-integrity process may well influence an outcome that affects a student's progression.

A defensible scoping record therefore states how flagged events are used. It should say who reviews them, whether a flag alone can trigger a sanction, and what the student can contest. It should apply the Commission's classification guidelines for providers and deployers, which include practical examples. If the institution relies on the derogation, the reasoning should be written down and reviewed whenever the configuration or use changes.

What changed on timing

Earlier planning assumed that high-risk provisions would apply on 2 August 2026. That premise is out of date. The Commission proposed the Digital Omnibus on AI on 19 November 2025, citing delays in standards for the high-risk rules and in designating national competent authorities and conformity assessment bodies. On 7 May 2026 the Council and Parliament announced a provisional agreement that fixes new dates. As of November 2026, Commission pages state that high-risk rules for areas including education apply from 2 December 2027.

The later date is a planning window, not a pause. Procurement, provider due diligence and evidence design take time, and contracts signed now will run past the application date. Re-check the date against the final published text before fixing programme milestones.

• 19 November 2025: Commission proposes the Digital Omnibus on AI.
• 7 May 2026: Council and Parliament announce a provisional agreement.
• 2 August 2026: Original application date for high-risk rules, now superseded.
• 2 December 2027: Stated application date for stand-alone high-risk systems, including education use cases.
• 2 August 2028: Stated date for high-risk systems embedded in products.

Provider and deployer obligations

Providers carry the heaviest product-level duties. Before placing a high-risk system on the EU market or putting it into service, a provider must complete a conformity assessment and implement a quality management system. It must also register the system in a public EU database. The assessed requirements include risk management, data quality, documentation and traceability, transparency, human oversight, accuracy, cybersecurity and robustness. The risk management system must be a continuous, iterative process across the lifecycle, covering known and reasonably foreseeable risks to health, safety or fundamental rights.

A university is typically the deployer. The Commission's summary states that before a high-risk system is deployed by entities providing public services, its fundamental rights impact must be assessed, and that high-risk systems and the entities using them must be registered in an EU database. Public universities should confirm with counsel whether these duties apply to them and what the registration entails. Deployers also need to assign ownership for ongoing review, because the Act treats risk management as lifecycle-wide.

In practice, the institution's leverage is in procurement. Ask the provider whether its documentation states monitoring and detecting prohibited behaviour as an intended purpose. Request its conformity assessment evidence, quality management system information and EU database registration details.

Where runtime policy enforcement fits

Some proctoring and exam-support workflows now involve AI models, agents and tool calls that touch student data. For those interactions, an enforcement point in the execution path can support human oversight and produce per-interaction evidence. Trussed enforces policy in the execution path across models, agents, tools and MCP servers: allow, redact, block, rate limit, or escalate for human approval. The scope is individual interactions and tool calls. Enforcement is applied at runtime, not reviewed after the fact, and it generates audit-ready evidence mapped to the EU AI Act, among other frameworks. Trussed supports education customers handling student data, and it can be self-managed in the customer's cloud with configurable retention and data residency.

This supports the controls above but does not replace them. No tool certifies compliance. Classification, the provider's conformity assessment, the fundamental rights impact assessment and legal review of GDPR questions remain with the institution and its providers.