However, being listed in Annex III does not mean that every AI system used in one of these areas is automatically subject to the full high-risk regime. Article 6 must be assessed together with Annex III and the applicable exceptions.
This guide explains how the Annex III classification works, the eight categories of high-risk use cases, the Article 6(3) exception, key provider and deployer obligations, and the current implementation timeline.
Important: The classification of an AI system depends on its intended purpose, functionality, deployment context, and other facts. This guide provides general information and should not be treated as a binding legal classification.
The EU AI Act establishes different regulatory requirements depending on the risk presented by an AI system.
The principal categories are:
For Annex III, the starting point is Article 6(2). An AI system falling within an Annex III use case is generally high-risk, subject to the Article 6(3) derogation and the other applicable provisions of the AI Act.
Article 6 provides two principal routes to high-risk classification.
An AI system may be high-risk where:
This route covers certain regulated product sectors, including areas such as medical devices, machinery, toys, lifts, radio equipment and other products covered by the relevant Union legislation.
An AI system may also be high-risk when it falls within one of the use cases listed in Annex III.
This is the route most relevant to many enterprise AI systems used in areas such as employment, education, credit, essential services, biometrics, law enforcement, migration, and justice.
Important: Classification should be based on the precise intended purpose and use case, not simply the industry in which the organisation operates.
Annex III identifies eight broad areas of high-risk AI use.
Relevant systems include certain:
The precise scope must be considered alongside the AI Act's prohibited practices and applicable Union or national law.
For example, biometric verification used solely to confirm that a person is who they claim to be is treated differently from remote biometric identification under Annex III.
This includes certain AI systems intended to be used as safety components in the management and operation of:
The system's actual function and intended use need to be assessed against the wording of Annex III.
Examples include AI systems used for:
An educational AI tool is therefore not automatically high-risk simply because it is used by a school or university. Its intended purpose must be assessed against the relevant Annex III use cases.
This category is particularly important for HR technology.
It covers specified AI systems used for activities such as:
AI used for employment-related decision-making should therefore be assessed carefully before deployment.
Annex III covers specified AI systems used in areas including:
Not every financial-services or public-sector AI system is automatically high-risk. The specific use case must fall within the relevant Annex III description.
Relevant use cases include certain AI systems used for:
This category must also be read alongside Article 5 and the specific restrictions applicable to certain law-enforcement AI practices.
Annex III includes specified systems used for:
Certain systems used for travel-document verification are treated differently under Annex III.
This category includes specified AI systems used to assist judicial authorities with activities such as:
It also covers certain AI systems intended to influence the outcome of elections or referenda.
The scope should be assessed against the exact wording of Annex III and the AI Act's prohibited-practice provisions.
No.
Article 6(3) provides a derogation for certain Annex III systems where the system does not pose a significant risk of harm to the health, safety, or fundamental rights of natural persons, including because it does not materially influence the outcome of decision-making.
The derogation can apply where the AI system is intended to:
However, the derogation does not apply where the AI system performs profiling of natural persons.
A provider relying on Article 6(3) must document its assessment before placing the system on the market or putting it into service and comply with the applicable registration requirement.
An AI system analyses job applications and ranks candidates for a company.
Potential classification: High-risk under the employment category in Annex III, depending on the precise intended purpose and functionality.
An HR system automatically schedules interviews but does not evaluate candidates or influence recruitment decisions.
Potential classification: The system may fall outside the relevant Annex III use case depending on its functionality and intended purpose.
An AI system evaluates an individual's creditworthiness for access to credit.
Potential classification: Potentially high-risk under Annex III, subject to the precise use case and applicable exceptions.
An AI system performs a narrow administrative task such as organising already-determined student records without influencing admission, assessment, or learning outcomes.
Potential classification: May fall outside the relevant high-risk use case, depending on the actual functionality.
These examples are illustrative only. A final classification should be based on the specific system, intended purpose, deployment environment, and applicable AI Act provisions.
Where an AI system is classified as high-risk and the relevant provisions apply, providers may need to address requirements including:
Establish and maintain an ongoing risk-management system covering the AI system's lifecycle.
Apply appropriate data-governance and data-management practices to relevant training, validation, and testing datasets.
Prepare and maintain the technical documentation required by the AI Act.
Maintain appropriate automatically generated logs and records.
Provide deployers with the information and instructions needed to use the system appropriately.
Design the system so that effective human oversight can be exercised.
Address appropriate levels of accuracy, robustness, and cybersecurity.
Establish the required quality-management processes.
Complete the applicable conformity-assessment procedure.
The exact conformity-assessment route depends on the classification and applicable legislation. It is therefore incorrect to assume that every Annex III system requires the same form of third-party assessment.
High-risk AI compliance is not limited to providers.
Deployers may have obligations relating to:
The obligations depend on the deployer's role, sector, use case, and the specific AI Act provisions applicable to the deployment.
Certain deployers of high-risk AI systems must conduct a Fundamental Rights Impact Assessment (FRIA) under Article 27.
The requirement is particularly relevant to:
Where applicable, the FRIA should consider matters such as:
A FRIA is distinct from a GDPR Data Protection Impact Assessment (DPIA), although the assessments may need to be coordinated where personal data is involved.
The AI Act does not replace the GDPR.
Where an AI system processes personal data, organisations may need to comply with both frameworks.
For example:
The two compliance programmes should therefore be coordinated rather than treated as interchangeable.
The implementation timeline has changed following the 2026 Digital Omnibus amendments.
Prohibited AI practices and AI literacy provisions began applying, subject to the specific provisions and amendments in force.
Rules concerning governance and general-purpose AI models began applying under the relevant provisions.
The AI Act's general application date and other provisions take effect, subject to the specific transition rules.
The high-risk requirements in Chapter III, Sections 1–3, apply to AI systems classified as high-risk under Article 6(2) and Annex III.
The corresponding high-risk requirements apply to AI systems classified under Article 6(1) and Annex I.
These dates are important because the previous 2 August 2026 date for Annex III high-risk obligations has been changed by Regulation (EU) 2026/1744.
Organisations should nevertheless begin preparation well before the legal application date, particularly where technical documentation, risk management, data governance, cybersecurity, and governance processes need to be established.
The AI Act provides different maximum administrative fines depending on the type of infringement.
For example, Article 99 provides for:
For undertakings, the applicable maximum is generally the higher of the specified fixed amount or the applicable percentage of worldwide annual turnover, subject to the AI Act's detailed rules.
These are statutory maximums, not automatic penalties. Actual enforcement outcomes depend on the nature and circumstances of the infringement and the applicable provisions.
Being a bank, university, hospital, or technology company does not automatically make every AI system high-risk.
The specific AI use case matters.
Article 6(3) provides a derogation for qualifying systems.
Classification depends heavily on what the AI system is intended to do and how it is actually designed and deployed.
Having a human somewhere in the workflow does not necessarily remove a system from high-risk classification. The Article 6(3) conditions must be assessed carefully.
The AI Act assigns different obligations to providers, deployers, importers, distributors, and other actors.
The current amended framework provides 2 December 2027 for Article 6(2)/Annex III high-risk systems.
No. Specific employment-related AI use cases are listed in Annex III. The actual intended purpose and functionality must be assessed.
Not necessarily. A general-purpose chatbot may be subject to transparency requirements rather than high-risk classification, but an AI system used for a specific Annex III decision-making use case can have a different classification.
No. The healthcare sector itself does not automatically determine Annex III classification. However, an AI system that is part of a regulated medical product may potentially fall under the Article 6(1)/Annex I route, which requires a separate assessment.
Potentially, yes. Article 6(3) provides a derogation where the required conditions are satisfied. The provider must document the assessment, and the profiling exclusion must also be considered.
Under the current amended Regulation, the Chapter III Sections 1–3 requirements apply to Article 6(2)/Annex III high-risk AI systems from 2 December 2027.
No. They address different regulatory requirements, although the frameworks can overlap where AI systems process personal data.
The Commission's classification guidance is intended to support consistent interpretation and enforcement, but the guidance itself is not a substitute for the AI Act's legal text. The Commission has published classification guidance and practical examples to help providers and deployers assess high-risk status.
Determining whether an AI system falls within Annex III is an important first step in building an AI governance programme.
RegulatoryBridge can help organisations assess AI system classifications, map applicable AI Act requirements, identify governance gaps, and develop implementation plans.
Assess your AI system with our AI Act Risk Classifier or speak with our AI Governance team about your requirements.