Risk Classification Framework

The AI Act's risk-based approach is its regulatory cornerstone. Classification determines whether your AI system is prohibited, subject to extensive requirements, requires transparency only, or faces no mandatory obligations. This chapter provides a systematic methodology for classification.

The Risk Classification Methodology

Classification follows a sequential assessment through four risk tiers:

Step 1: Is it Prohibited? (Article 5)

First, determine if the AI practice falls under the eight prohibited categories. If yes, the AI cannot be legally deployed in the EU—stop here.

Step 2: Is it High-Risk? (Article 6)

If not prohibited, assess whether the AI meets either high-risk pathway:

• Pathway A: Safety component in Annex I product requiring third-party conformity assessment
• Pathway B: Standalone AI in an Annex III use case (subject to "filter" exception)

Step 3: Does it Require Transparency? (Article 50)

If not high-risk, check transparency triggers:

• Direct human interaction (chatbots)
• Synthetic content generation (deepfakes)
• Emotion recognition or biometric categorisation (permitted contexts)

Step 4: Minimal Risk

If none of the above apply, the system is minimal risk with no mandatory requirements.

High-Risk Pathway A: Annex I Product Safety (Article 6(1))

AI systems are high-risk under Article 6(1) if they meet BOTH conditions:

1. The AI is a safety component of a product, or is itself a product, covered by EU harmonisation legislation listed in Annex I
2. The product requires third-party conformity assessment under that legislation

Annex I Sectors (Product Safety Legislation)

High-Risk Pathway B: Annex III Use Cases (Article 6(2))

AI systems are high-risk under Article 6(2) if they fall within the use cases enumerated in Annex III—regardless of the product in which they are deployed.

Annex III: The Eight High-Risk Domains

Section 1: Biometrics (where permitted)

• Remote biometric identification systems
• Biometric categorisation systems
• Emotion recognition systems (non-prohibited contexts)

Section 2: Critical Infrastructure

• AI safety components in management/operation of:
  • Road traffic
  • Water, gas, heating, electricity supply
  • Digital infrastructure

Section 3: Education and Vocational Training

• AI determining access to education institutions
• AI evaluating learning outcomes
• AI assessing appropriate education level
• AI monitoring prohibited behaviour during tests

Section 4: Employment, Workers Management, Self-Employment Access

• Recruitment and selection (CV screening, interviews)
• Promotion, termination, task allocation decisions
• Performance and behaviour monitoring

Section 5: Access to Essential Services

• Evaluation of eligibility for essential public assistance benefits and services, including healthcare (by public authorities)
• Credit worthiness evaluation (individuals)
• Life and health insurance risk assessment (individuals)
• Emergency services dispatch prioritisation

Section 6: Law Enforcement (where permitted)

• Victim risk assessment (risk of becoming victim of criminal offences)
• Polygraphs and similar tools
• Evidence reliability assessment
• Offence risk assessment (profiling exception)

Section 7: Migration, Asylum, Border Control

• Polygraphs and similar tools
• Immigration/asylum/visa application risk assessment
• Document authenticity verification
• Visa/permit/complaint examination assistance

Section 8: Administration of Justice and Democracy

• Judicial fact and law research assistance
• AI with potential to influence electoral outcomes

The "Filter" Exception (Article 6(3))

A critical nuance: Annex III AI systems are NOT automatically high-risk. Article 6(3) provides a "filter" exception:

An Annex III AI system is not high-risk if it does not pose a significant risk of harm to health, safety, or fundamental rights, taking into account:

⚠️ Compliance Warning: The filter exception is narrow and must be justified. Providers must document their filter analysis. If uncertain, treat the system as high-risk.

⚠️ Art. 6(4) Registration Obligation: Providers who consider an Annex III system is NOT high-risk must (a) document their assessment before placing on market or putting into service, and (b) register under Article 49(2). This is a mandatory obligation — providers invoking Art. 6(3) must register, not merely document. Documentation must be made available to national competent authorities on request.

Filter Exception Does NOT Apply To:

• AI systems performing profiling of natural persons (always high-risk per Article 6(3))
• AI used for profiling of natural persons

Commission Guidelines on Classification (Article 6(5))

Article 6(5) required the Commission to publish guidelines specifying the practical implementation of Article 6, including a comprehensive list of high-risk and not-high-risk use case examples, by 2 February 2026. These guidelines (once published) are a key reference for applying the Art. 6(3) filter exception and should be consulted when making classification determinations. Providers should monitor the Official Journal and the European AI Office for publication of these guidelines.

Limited Risk: Transparency Obligations (Article 50)

AI systems that are not prohibited or high-risk may still have transparency obligations:

Minimal Risk: Voluntary Measures

The majority of AI systems fall into minimal risk with no mandatory requirements:

• Spam filters
• AI-enabled video games
• Inventory management AI
• General recommendation systems

Providers may voluntarily adopt codes of conduct (Article 95) applying high-risk requirements.

Risk Classification Quick Assessment

Use this checklist for preliminary classification: