EU AI Act Risk Classification Explained

Key Takeaways

Point	Summary
Risk tiers	Unacceptable, high-risk, limited risk, and minimal risk
Unacceptable risk	Prohibited AI practices that threaten fundamental rights or safety
High-risk	Systems subject to extensive requirements before market placement
Limited risk	Systems requiring transparency obligations (e.g., disclosure of AI use)
Minimal risk	No specific requirements, but codes of conduct encouraged

Quick Answer: The EU AI Act classifies AI systems into four risk categories. Most AI falls into minimal risk with no requirements. High-risk systems face extensive compliance obligations, while unacceptable-risk AI is prohibited entirely. Your classification depends on the AI system's use case and potential impact.

The Four Risk Categories

1. Unacceptable Risk (Prohibited)

Certain AI applications are banned outright because they pose clear threats to people's safety, livelihoods, or rights.

Prohibited practices include:

Practice	Description
Social scoring	Public authority systems that evaluate individuals based on social behavior leading to unfavorable treatment
Subliminal manipulation	Techniques that deploy subliminal elements to materially distort behavior in harmful ways
Exploitation of vulnerabilities	Systems that exploit vulnerabilities of specific groups (age, disability, economic situation)
Real-time biometric identification	Remote identification in public spaces for law enforcement (with narrow exceptions)
Emotion recognition	Recognition of emotions in workplaces and educational institutions (with exceptions)
Biometric categorization	Categorizing individuals based on biometric data to infer sensitive attributes
Untargeted facial scraping	Creating facial recognition databases through untargeted scraping
Predictive policing	Making predictions about criminal offense risk based solely on profiling

Exceptions exist for:

• Real-time biometric identification may be permitted for specific law enforcement purposes (missing children, imminent threats, serious crimes) with judicial authorization
• Emotion recognition for medical or safety purposes may be allowed

2. High-Risk AI Systems

High-risk AI systems are permitted but subject to extensive requirements. A system is classified as high-risk if it falls into one of two categories:

Category 1: Safety Components

AI systems that are:

• Safety components of products covered by EU harmonization legislation (e.g., medical devices, machinery, toys)
• Products themselves that require third-party conformity assessment

Category 2: Annex III Listed Uses

AI systems used in specific high-risk areas:

Area	Examples
Biometrics	Remote biometric identification, biometric categorization, emotion recognition
Critical infrastructure	Management of road traffic, water, gas, heating, electricity
Education	Determining access to education, evaluating learning outcomes, detecting cheating
Employment	CV screening, interview analysis, promotion decisions, task allocation, termination
Essential services	Credit scoring, life/health insurance pricing, emergency services dispatching
Law enforcement	Evidence reliability assessment, polygraphs, crime analytics, profiling
Migration and asylum	Risk assessments, document authenticity verification
Justice	Assisting judicial authorities in researching and interpreting facts and law

Important exception: An AI system listed in Annex III is NOT considered high-risk if it does not pose a significant risk of harm to health, safety, or fundamental rights, including by not materially influencing decision-making outcomes.

This exception requires the provider to document the assessment and register it before placing the system on the market.

3. Limited Risk AI Systems

Limited risk systems have transparency obligations but fewer technical requirements than high-risk systems.

System Type	Transparency Requirement
AI interacting with humans	Users must be informed they are interacting with an AI system
Emotion recognition systems	Users must be informed that such a system is in use
Biometric categorization	Users must be informed of the categorization
AI-generated content	Must be marked as artificially generated or manipulated (deep fakes)

4. Minimal Risk AI Systems

The majority of AI systems fall into this category and face no specific legal requirements under the EU AI Act.

Examples of minimal risk AI:

• Spam filters
• AI-enabled video games
• Recommendation algorithms for entertainment
• AI-assisted writing tools
• Inventory management systems

While not legally required, the regulation encourages providers of minimal risk systems to voluntarily adopt codes of conduct that apply high-risk requirements.

How to Classify Your AI System

Follow this decision process to determine your AI system's risk classification:

Step 1: Check for Prohibited Practices

Does your AI system involve any of the explicitly prohibited uses?

• If yes: The system cannot be placed on the EU market
• If no: Continue to Step 2

Step 2: Check for Safety Component Status

Is your AI system a safety component of a product covered by EU product safety legislation, or is it such a product itself?

• If yes and third-party conformity assessment is required: High-risk
• If no: Continue to Step 3

Step 3: Check Annex III Categories

Is your AI system used for any purpose listed in Annex III (biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, justice)?

• If yes: Likely high-risk, unless the exception applies
• If no: Continue to Step 4

Step 4: Check for General-Purpose AI

Is your AI system a general-purpose AI model (GPAI)?

• If yes: Specific GPAI rules apply (separate framework)
• If no: Continue to Step 5

Step 5: Check Transparency Obligations

Does your system interact directly with natural persons, generate synthetic content, or perform emotion recognition?

• If yes: Limited risk with transparency obligations
• If no: Minimal risk with no specific requirements

High-Risk AI: The Exception That May Apply

Even if your AI system falls under Annex III categories, you may be able to demonstrate it is not high-risk if:

1. The system performs a narrow procedural task
2. The system improves the result of a previously completed human activity
3. The system detects decision-making patterns without replacing human assessment
4. The system performs a preparatory task for an assessment

Documentation requirements for claiming the exception:

• Written assessment explaining why the system does not pose significant risks
• Risk assessment methodology used
• Evidence supporting the determination
• Registration in the EU database before market placement

This exception is narrow and should be applied carefully. When in doubt, assume high-risk classification applies.

General-Purpose AI Models (GPAI)

The EU AI Act includes specific rules for general-purpose AI models, recognizing that these powerful foundation models can be used for many purposes:

GPAI Category	Requirements
All GPAI models	Technical documentation, copyright compliance information, EU representative designation
GPAI with systemic risk	Additional obligations including model evaluations, adversarial testing, cybersecurity measures, incident reporting

A GPAI model is presumed to have systemic risk if:

• It has high impact capabilities, presumed when cumulative training compute exceeds 10^25 FLOPs. This is a rebuttable presumption; providers may contest with evidence that the model does not pose systemic risk.
• The European Commission designates it based on specific criteria

Note: The 10^25 FLOP threshold may be amended by the Commission via delegated acts to reflect technological developments. Providers must notify the Commission within two weeks of reasonably foreseeing or reaching this threshold, even before placing the model on the market.

Practical Examples

AI System	Classification	Reasoning
Resume screening tool	High-risk	Employment use case in Annex III
Customer service chatbot	Limited risk	Requires disclosure of AI interaction
Email spam filter	Minimal risk	No significant harm potential
Credit scoring algorithm	High-risk	Essential services use case in Annex III
AI-generated marketing images	Limited risk	Synthetic content requires marking
Predictive maintenance	Generally minimal	Unless critical infrastructure safety component
Medical diagnosis assistant	High-risk	Medical device safety component
Recommendation engine (retail)	Minimal risk	Entertainment/commerce, no significant harm

What Happens If You Misclassify?

Misclassifying an AI system to avoid requirements carries significant risks:

• Penalties up to 15 million EUR or 3% of global turnover for providing incorrect information to authorities
• Market access denied if the misclassification is discovered during market surveillance
• Liability exposure if harms occur from an AI system that should have had safeguards
• Reputational damage from regulatory enforcement actions

How Bastion Helps

Bastion helps organizations navigate AI Act classification:

• AI system inventory. We help identify and catalog all AI systems across your organization.
• Classification assessment. We evaluate each system against EU AI Act criteria and document the risk classification.
• Gap analysis. For high-risk systems, we identify what compliance measures are needed.
• Documentation support. We help prepare the assessments and documentation required for claiming exceptions where applicable.
• Ongoing monitoring. As guidance evolves and the EU issues clarifications, we help ensure classifications remain accurate.

---

Ready to classify your AI systems? Talk to our team

---

Sources

• EU AI Act Annex I and III (EUR-Lex) - Official list of high-risk AI system categories
• European Commission AI Risk Classification - Overview of the risk-based approach