Cyber Cultura course
ISO 23894: AI Risk Management
AI risks across the entire lifecycle of data and models.
Enables participants to establish and operate an AI risk management process, aligned with ISO/IEC 23894 and the principles of ISO 31000, covering the entire lifecycle of data and models.
- Workload
- 5 hours
- Format
- Online or in-company
- Investment
- On request
We’ll reply with dates, format and investment.
Who it’s for
- Risk managers who are now responsible for AI systems.
- Data and AI teams that need to document and treat risks.
- Compliance, privacy and security professionals involved in AI projects.
What you’ll develop
- Understand the risk principles and framework applied to AI systems.
- Carry out the identification, analysis, evaluation and treatment of AI risks.
- Address sociotechnical risks: bias, fairness, interpretability, robustness, security, privacy and functional safety.
- Define metrics, thresholds, monitoring and review triggers.
- Integrate AI risk into changes, incidents and third parties.
Syllabus
5 hours
01Fundamentals
- Risk management principles (ISO 31000) in the AI context
- AI risk terminology: sources, events, impacts and uncertainties
02Context establishment
- Organizational, regulatory and usage context
- Risk criteria, appetite and tolerance for AI
- Scope, boundaries and interfaces of the AI system
03The risk process across the lifecycle
- Identification: data, models, pipelines, people, suppliers and the operating environment
- Analysis: likelihood, impact and severity; dependencies and cascading effects
- Evaluation: acceptance criteria and prioritization
- Treatment: avoid, reduce, share or accept; technical and organizational safeguards
04AI-specific risks
- Data: quality, representativeness, bias, privacy and security
- Model: overfitting, drift, robustness against attacks (evasion and poisoning) and interpretability
- Use and context: human error, overautomation, oversight and social and ethical effects
- Security and privacy: attacks on models, membership inference and differential privacy
05Measurement and monitoring
- Performance, robustness, fairness and explainability metrics
- Drift, data and model monitoring, operating thresholds and action triggers
- Logging, traceability and evidence
06Communication and consultation
- Interested parties and transparency proportional to risk
- Artifacts: AI risk by use case, an AI impact assessment (AIIA) where applicable, and executive reports
07Integrations
- Relationship with the AIMS (ISO/IEC 42001), security (27001), privacy (27701), continuity (22301) and NIST AI RMF
- Risk in changes, incidents and third parties
What the course includes
- AI risk artifact templates
For your team
This course is also offered in-company, on-site or remote, as a private class for your company.
Cyber Cultura course
ISO 23894: AI Risk Management
We’ll reply with dates, format and investment.
cyber@cyberconsulting.com.br