This document is under active development and has not been finalised.
Skip to content

Risk Management – Art. 9

Requirement

Providers of high-risk AI systems must implement a risk management system as a continuous iterative process throughout the entire lifecycle.

Mandatory Components

The risk management system must:

  1. Identify and analyse known and foreseeable risks — during intended use and during reasonably foreseeable misuse
  2. Assess risks that may arise during intended use and during reasonably foreseeable misuse
  3. Develop and implement appropriate risk mitigation measures
  4. Assess and document residual risks — taking into account the risk mitigation measures
  5. Test — before placing on the market and throughout the lifecycle, to ensure the system functions as intended

Testing Requirements

  • Tests against pre-defined metrics and thresholds
  • Consideration of the intended purpose
  • Testing for bias with regard to affected groups of persons
  • Regardless of data format: structured, unstructured, mixed

BAUER GROUP Implementation (Minimal Approach)

MeasureImplementation
Risk identificationExtension of the existing CRA risk assessment form with AI-specific risk categories
Risk assessmentSimple risk matrix (likelihood × impact)
Risk mitigationDocumentation of chosen measures in the CRA risk assessment form
TestingIntegration into the existing CI/CD pipeline (ML model metrics as quality gate)
DocumentationTemplate-based, product-specific

Synergies with CRA

CRA vulnerability management already covers cybersecurity risks. AI Act risk management supplements this with AI-specific risks (bias, hallucination, explainability, robustness).

Documentation licensed under CC BY-NC 4.0 · Code licensed under MIT