High-Risk AI System

An AI system is automatically classified as high-risk if it serves as a safety component of a product covered by specific EU harmonization legislation listed in Annex I, or if the AI system itself is that product. This includes machinery, medical devices, lifts, toys, and radio equipment. The system must be intended to perform a safety function that directly controls or influences the product's safe operation, where a failure or malfunction could directly endanger human health or property.

A standalone AI system is classified as high-risk if it operates in one of eight critical domains listed in Annex III, where its application creates a significant risk of harm. These domains include:

• Biometric identification and categorization of natural persons
• Management and operation of critical infrastructure
• Education and vocational training
• Employment, worker management, and access to self-employment
• Access to essential private and public services (e.g., credit scoring, insurance)
• Law enforcement
• Migration, asylum, and border control management
• Administration of justice and democratic processes

An AI system is high-risk when it performs automated profiling of individuals to make or inform a consequential decision that has a legal effect or similarly significant impact. This includes decisions about a person's legal status, financial creditworthiness, employment eligibility, or access to essential services. The key trigger is the combination of automated personality or behavior analysis with a materially impactful outcome, which demands strict oversight to prevent discrimination and protect fundamental rights.

Beyond physical safety, a system is high-risk if its intended purpose creates a foreseeable risk to fundamental rights as enshrined in the EU Charter. This includes rights to non-discrimination, data protection, privacy, freedom of expression, and a fair trial. A mandatory Fundamental Rights Impact Assessment (FRIA) must be conducted by the deployer to document and mitigate these specific risks, making the potential for rights infringement a core classification characteristic.

A defining characteristic of a high-risk system is the legal requirement for meaningful human oversight. The system must be designed with appropriate human-machine interface tools allowing natural persons to fully understand its capabilities and limitations, monitor its operation for anomalies, and intervene or override its decisions. This oversight must be effective, not a token gesture, ensuring a human can decide to disregard the AI's output to prevent or minimize risks.

Systems that identify or categorize individuals based on biological or behavioral characteristics.

• Remote biometric identification in public spaces (law enforcement exceptions apply)
• Emotion recognition in workplaces or educational institutions
• Biometric categorization inferring race, political opinions, or religious beliefs

These systems pose direct risks to privacy rights and non-discrimination principles under GDPR and the EU Charter.

AI systems used as safety components in the management and operation of critical digital and physical infrastructure.

• Road traffic management and autonomous driving systems
• Water, gas, heating, and electricity supply network controls
• Telecommunications network routing and load balancing

Failure in these systems can cause cascading societal disruption and direct physical harm to populations.

Systems determining access, admission, or evaluation within educational institutions.

• Automated exam scoring for high-stakes university admissions
• Student placement algorithms assigning pupils to educational tracks
• Proctoring systems detecting cheating through gaze or keystroke analysis

These systems can permanently alter life trajectories through biased assessments or unfair access barriers.

AI used for recruitment, promotion, task allocation, or performance evaluation.

• CV screening and candidate ranking tools for hiring decisions
• Algorithmic task assignment in gig economy platforms
• Performance monitoring systems tracking keystrokes, breaks, or productivity

These applications directly affect labor rights, equal opportunity, and worker dignity under EU social law.

Systems evaluating eligibility for public benefits, financial services, or insurance.

• Creditworthiness assessment for mortgage or loan applications
• Risk pricing in health and life insurance underwriting
• Prioritization of emergency service dispatch

These systems determine access to essential services and can perpetuate systemic discrimination against protected groups.

AI systems used by public authorities for policing, border control, and judicial processes.

• Individual risk assessments predicting recidivism or flight risk
• Evidence reliability evaluation tools in criminal proceedings
• Migration and asylum application triage systems
• Deepfake detection for forensic analysis

These systems engage due process rights, presumption of innocence, and freedom of movement.

An independent, accredited organization designated by an EU member state to conduct third-party conformity assessments. These bodies must demonstrate impartiality, technical competence, and strict confidentiality. Their accreditation scope defines which categories of AI systems they are authorized to evaluate. They serve as the critical gatekeepers ensuring high-risk systems meet harmonized standards.

A mandatory documented process for deployers of high-risk AI systems to evaluate specific risks to individuals' rights. Key components include:

• A detailed description of the intended purpose and deployment context
• An analysis of risks to vulnerable groups
• Mitigation measures and human oversight protocols
• The timeframe and frequency of the system's use

A continuous, systematic process where providers collect and analyze real-world performance data from deployed high-risk AI systems. This is not a one-time event but an ongoing obligation to identify emerging risks, drift, and unintended consequences. Findings must feed back into the risk management system and can trigger serious incident reporting to market surveillance authorities.

The comprehensive dossier demonstrating a high-risk AI system's compliance. It must contain:

• A detailed description of the system's design and architecture
• Specifications for training, validation, and testing datasets
• Performance metrics and accuracy benchmarks
• Human oversight measures and logging mechanisms This document is subject to audit by notified bodies and market surveillance authorities.

A mandatory obligation requiring providers to immediately notify market surveillance authorities of any malfunction or failure causing death, serious health damage, or significant property damage. Reports must include a root cause analysis, corrective actions taken, and measures to prevent recurrence. This mechanism enables rapid regulatory response to systemic risks.