DataTerminology.comNEW
Practices to identify, assess, prioritize, treat, and monitor risks to people, organizations, and society arising from AI systems.
DESCRIPTION
AI risk management tailors classical risk disciplines to AI’s socio-technical nature. It addresses harms from bias, privacy leakage, security threats, robustness failures, and misuse. Processes include risk identification, likelihood-impact evaluation, control selection, testing, and continuous monitoring. Frameworks like NIST AI RMF and ISO/IEC 23894 guide governance, measurement, and management functions. Effective programs integrate with MLOps for automated checks and incident handling. Documentation and human oversight enhance accountability and transparency.
ALTERNATIVE DEFINITIONS
ISO/IEC 23894
This document provides guidance on risk management for artificial intelligence (AI).
Positions AI risk management as guidance aligned to general risk standards (e.g., ISO 31000).
NIST
The AI RMF is intended to help organizations manage risks to individuals, organizations, and society associated with artificial intelligence.
Emphasizes multi-stakeholder risks and organizational adoption of risk practices.
RELATED TERMS
NIST AI Risk Management Framework ISO-IEC 23894 AI Governance Impact Assessment Monitoring Incident Response Systematic Error Security
COMPETING TERMS
AI Safety Model Risk Management
RELATIONS
- Broader: Risk Management
- Conforms to: ISO-IEC 23894 NIST AI Risk Management Framework
KNOWLEDGE AREAS
AI Governance MLOps AI Data Management
TARGET AUDIENCE
Chief Data Officer AI Product Manager MLOps Engineer Data Governance Manager Chief Technology Officer Data Scientist Enterprise Architect
EXTERNAL SOURCES
- Wikipedia - AI safety
- NIST - Official AI RMF overview
- NIST - AI RMF Playbook guidance
- ISO/IEC 23894 - AI risk management guidance
- European Commission - EU AI Act overview and risks
TOP QUESTIONS
What types of risks does AI Risk Management address?
It covers safety, security, privacy, bias, robustness, compliance, and societal harms, along with operational and reputational risks.
Category: Overview
How does AI Risk Management integrate with MLOps?
Risk controls become automated checks in pipelines, with continuous monitoring, drift alerts, incident workflows, and approval gates.
Category: Integration
What are common AI Risk Management challenges?
Defining risk metrics, limited ground truth for fairness, dynamic adversaries, and aligning cross-functional stakeholders are frequent hurdles.
Category: Challenges
Which standards guide AI Risk Management?
NIST AI RMF and ISO/IEC 23894 provide governance, measurement, and management guidance tailored to AI systems.
Category: Standards
How do you prioritize AI risks?
Use impact–likelihood matrices, risk appetite thresholds, and context factors like affected populations and misuse potential.
Category: HowTo
What artifacts support AI Risk Management?
Risk registers, impact assessments, model and system cards, and audit logs provide traceability and accountability.
Category: Artifacts
How is AI Risk Management different from traditional risk management?
It considers socio-technical contexts, human factors, data issues, and non-deterministic model behaviors in addition to standard controls.
Category: Comparison
What metrics are used in AI Risk Management?
Fairness, robustness, privacy leakage, calibration, and monitoring SLIs/SLOs quantify residual risk and control effectiveness.
Category: Metrics
When should AI Risk Management begin?
At ideation, continuing through design, development, deployment, and operations with periodic re-assessment as context changes.
Category: Lifecycle