Risk Management System (Article 9)

Learning Objectives

By the end of this chapter, you will be able to:

Design and implement a compliant AI risk management system
Apply the Article 9 iterative risk management process
Identify, analyse, and evaluate AI-specific risks
Select and implement appropriate risk mitigation measures
Integrate AI risk management with existing enterprise frameworks

Article 9 establishes the foundational compliance requirement for high-risk AI: a comprehensive, continuous risk management system. This is not a one-time assessment but an iterative process spanning the entire AI lifecycle—from conception through decommissioning.

The Article 9 Risk Management Framework

Article 9(2) Iterative Process

The AI Act requires a continuous iterative process comprising four steps as set out in Article 9(2)(a)-(d):

Step	Article 9(2)	Activities	Timing
(a) Identification and analysis	9(2)(a)	Identify and analyse the known and the reasonably foreseeable risks that the high-risk AI system can pose to health, safety, or fundamental rights	Pre-development, ongoing
(b) Estimation and evaluation	9(2)(b)	Estimate and evaluate the risks that may emerge when the high-risk AI system is used in accordance with its intended purpose and under conditions of reasonably foreseeable misuse	Pre-market, post-market
(c) Evaluation of post-market data	9(2)(c)	Evaluate other risks that may emerge based on the analysis of data gathered from the post-market monitoring system referred to in Article 72	Post-deployment, continuous
(d) Adoption of risk management measures	9(2)(d)	Adopt appropriate and targeted risk management measures designed to address the risks identified pursuant to point (a)	Throughout lifecycle

Lifecycle Coverage Requirement

Article 9(1) mandates risk management "throughout the entire lifecycle":

Pre-Development Phase:

Intended use and context analysis
Initial risk identification
Risk management strategy design

Development Phase:

Training data risk assessment
Model behaviour analysis
Bias and fairness evaluation

Testing Phase:

Validation and verification risks
Performance boundary testing
Foreseeable misuse scenarios

Deployment Phase:

Real-world risk monitoring
Human oversight integration
Incident response planning

Post-Market Phase:

Continuous performance monitoring
Incident data analysis
Risk re-evaluation cycles

Types of Risks to Address

Health and Safety Risks (Article 9(2)(a))

Physical and mental health impacts including:

Direct harm from AI decisions (e.g., medical diagnosis errors)
Indirect harm from system failures (e.g., autonomous vehicle accidents)
Psychological impacts (e.g., unfair denial of services)

Fundamental Rights Risks (Article 9(2)(b))

Impacts on EU Charter rights including:

Non-discrimination (Article 21): Algorithmic bias
Privacy (Article 7): Excessive data collection
Fair trial (Article 47): Opaque decision-making
Human dignity (Article 1): Dehumanising treatment
Freedom of expression (Article 11): Content moderation bias

Contextual Risk Factors

The risk management system must consider a range of factors drawn from multiple paragraphs of Article 9. Article 9(4) requires that when assessing risk management measures, due consideration shall be given to the effects and possible interactions resulting from the combined application of the requirements set out in Chapter III, Section 2, with a view to minimising risks more effectively while achieving an appropriate balance in implementing the measures to fulfil those requirements. Additionally, Article 9(9) requires that risk management measures give particular consideration to whether the high-risk AI system is likely to be accessed by or have an impact on persons under the age of 18 or, as appropriate, other vulnerable groups.

Factor	Source	Risk Consideration
Intended purpose	Art. 9(2)(a)-(b)	What harm could occur during normal use?
Foreseeable misuse	Art. 9(2)(b)	What if users misuse the system?
System interactions	Art. 9(4)	Effects and possible interactions from combined requirements
Vulnerable groups and persons under 18	Art. 9(9)	Are children, disabled, elderly especially affected?
Cumulative effects	Art. 9(2)	What happens with repeated decisions over time?
Bias potential	Art. 10(2)(f)-(g)	Could training data or design introduce discrimination?
Operational environment	Art. 9(2)(a)-(b)	What contextual factors affect risk?

Compliance Note

You must assess risks for BOTH intended use AND reasonably foreseeable misuse. Many compliance failures stem from ignoring misuse scenarios.

Risk Identification Methods

Systematic Risk Identification Techniques

Method	Description	When to Use
HAZOP Analysis	Structured deviation analysis	Complex AI systems
FMEA	Failure modes and effects analysis	Component-level risks
Threat Modelling	Adversarial scenario analysis	Security risks
Bias Audits	Fairness assessment across groups	Discrimination risks
Red Teaming	Adversarial testing by experts	Pre-deployment
Stakeholder Consultation	Input from affected groups	Rights impact assessment

AI-Specific Risk Categories

Technical Risks:

Model accuracy degradation
Data drift and distribution shift
Adversarial vulnerabilities
Unexplainable outputs

Operational Risks:

Human oversight failures
Integration errors
Misinterpretation of outputs
Automation bias

Ethical Risks:

Discriminatory outcomes
Privacy violations
Lack of transparency
Accountability gaps

Risk Evaluation Criteria

Risk Prioritisation Matrix

Evaluate each identified risk using severity and likelihood:

Likelihood / Severity	Negligible	Minor	Moderate	Major	Critical
Almost Certain	Medium	High	High	Critical	Critical
Likely	Low	Medium	High	High	Critical
Possible	Low	Medium	Medium	High	High
Unlikely	Low	Low	Medium	Medium	High
Rare	Low	Low	Low	Medium	Medium

Acceptability Thresholds

Article 9(5) requires that residual risk be acceptable considering:

State of the art in risk mitigation
Benefits of the AI system
Generally acknowledged standards
Expectations of intended users

Risk Mitigation Measures

Mitigation Strategy Hierarchy

Apply measures in priority order (Article 9(5)):

1. Elimination (preferred)

Remove the risk source entirely
Redesign to avoid risk scenarios

2. Reduction

Technical safeguards
Operational controls
User training

3. Transfer

Human-in-the-loop for high-stakes decisions
Insurance coverage
Contractual risk allocation

4. Acceptance (last resort)

Residual risk deemed acceptable
Must be documented and justified

Common AI Risk Mitigation Measures

Risk Type	Mitigation Measures
Accuracy risks	Validation testing, confidence thresholds, fallback rules
Bias risks	Diverse training data, bias testing, fairness constraints
Robustness risks	Adversarial training, input validation, anomaly detection
Transparency risks	Explainability methods, audit trails, user disclosure
Oversight risks	Human review triggers, override mechanisms, escalation paths

Documentation Requirements

Risk Management Documentation

Your risk management system must document:

Element	Required Content
Risk register	All identified risks with classifications
Assessment methodology	How risks were identified and evaluated
Mitigation measures	Actions taken and their effectiveness
Residual risks	Remaining risks after mitigation
Review schedule	When re-assessments occur
Responsibilities	Who is accountable for risk management

Expert Insight

Integrate AI risk management with your existing enterprise risk management (ERM) framework. This ensures consistency, leverages existing governance, and avoids duplication.

Integration with Other Requirements

Relationship to Other Article 9 Elements

Requirement	Integration Point
Data Governance (Art. 10)	Data quality risks feed into risk assessment
Technical Documentation (Art. 11)	Risk documentation is mandatory content
Logging (Art. 12)	Logs enable post-market risk monitoring
Human Oversight (Art. 14)	Oversight is a key risk mitigation measure
Post-Market Monitoring (Art. 72)	Continuous risk re-evaluation

Risk Management Compliance Checklist

Risk management system established and documented
Lifecycle-spanning process implemented
All risk types identified (health, safety, rights)
Foreseeable misuse scenarios assessed
Vulnerable group impacts considered
Risk evaluation criteria defined
Mitigation measures implemented
Residual risks documented and justified
Regular review cycle established
Integration with ERM complete

Risk Management System