Domain 3 of 4 · Chapter 2 of 3

Establish AI Governance and Regulatory Compliance

Unlock the complete study guide.

Bundled into the existing AWS Certified AI Business Strategist premium course — no separate purchase.

14-day refund guarantee on one-time purchases and your first subscription charge.

Included in this chapter:

  • AI governance structures and regulatory compliance
  • Accountability: the AI governance council and owners
  • The AI inventory and shadow AI
  • Extending enterprise controls to AI, suppliers included
  • Rating AI risk: likelihood, severity, and context
  • From risk rating to governance decision
  • Keeping AI risk ratings current
  • Mapping legal obligations to the actual use
  • Privacy obligations along the AI data flow
  • The compliance evidence record
  • Shared responsibility for managed AI services
  • Access: identity, authorization, and least privilege
  • Protecting information by sensitivity class
  • AI-specific threats: prompt injection and data poisoning
  • Choosing among governance standards and frameworks
  • NIST AI RMF functions
  • The NIST Generative AI Profile (NIST AI 600-1)
  • Exam-pattern recognition for AI governance

Four building blocks of AI governance, four different classifications

AspectAccountable structureRisk classificationLegal obligationsProtective controls
Question it answersWho may decide, and who owns each task?How much scrutiny does this use need?What does the law require for this use?How are data and access protected?
What gets classifiedEach AI tool or use: approved, blocked, or under evaluationEach potential harm of a specific use: likelihood and severity combine into a risk ratingThe use: its legal category, such as prohibited or high-riskInformation: its sensitivity class
Who decidesThe AI governance council, with named ownersAssessors applying defined criteria; the AI governance council accepts residual riskLegal and compliance specialistsSecurity, compliance, and business teams through a classification policy
Common substitution errorTreating tool approval as approval of every use and data typeTreating a low score as a change in legal categoryTreating a supplier certificate as the organization's own complianceTreating strong sign-in as permission for broad access

Decision tree

Practice prohibitedby law?YesDo not deployNoLegal high-riskcategory?YesStatutoryduties andrigorous reviewNo; other duties, such astransparency, may applyHigh internalrisk rating?YesRigorousreviewNoLighterdocumented pathAlways: residual risk beyond tolerancemeans do not deploy or cease safely;reclassify when the use or itscontext changes

Cheat sheet

  • AI governance needs cross-functional representation and explicit decision authority
  • Shared AI work still needs named owners and clear handoffs
  • A centralized AI inventory should make ownership and permitted use visible
  • Managing shadow AI requires a usable approved path as well as policy enforcement
  • AI governance should extend existing enterprise controls rather than operate in isolation
  • Buying AI changes the delivery arrangement but does not remove the need for supplier governance
  • AI compliance depends on where the system operates and how the organization uses it
  • A privacy impact assessment examines the proposed processing before risks become embedded
  • Permission to hold customer information is not blanket permission to reuse it for AI
  • AI-generated information about an identifiable person can carry privacy obligations even when it is wrong
  • AI vendor review must follow data beyond the company's immediate interface
  • Compliance evidence should connect obligations, assessments, controls, and approval decisions
  • Legal AI risk categories impose duties that an internal risk score cannot remove
  • AI access should be limited to the data and functions each role needs
  • Strong identity controls and appropriate permissions solve different access problems
  • Encryption protects stored and transmitted AI data but must be combined with access controls
  • Sensitive AI information should be minimized before use and screened for inappropriate disclosure
  • Monitoring records need their own access, retention, and privacy controls
  • AWS infrastructure protection does not replace the customer's AI data responsibilities
  • Prompt injection can turn untrusted material into attempted instructions
  • Data poisoning targets information the AI system learns from or relies on
  • Information sensitivity determines the handling rules for AI use
  • Risk classification combines likelihood with the severity and reversibility of harm
  • The business use and degree of autonomy shape an AI system's risk
  • Risk remaining after controls must be assessed against the organization's tolerance
  • Governance effort should be proportional to risk and able to stop unacceptable uses
  • An AI risk register should track harms and treatment evidence across use cases
  • AI risk ratings require review as the system and its operating context evolve
  • ISO/IEC 42001 provides requirements for an organizational AI management system
  • ISO/IEC 23053 provides a common framework for describing AI systems that use machine learning
  • The NIST AI RMF connects Govern, Map, Measure, and Manage rather than treating risk as a one-time checklist
  • AWS CAF for AI places responsible use within a broader organizational adoption framework
  • The NIST Generative AI Profile applies the AI RMF to risks amplified by generative AI

Unlock with Premium — includes all practice exams and the complete study guide.

Also tested in

References

  1. https://docs.aws.amazon.com/aws-certification/latest/ai-business-strategist-01/ai-business-strategist-01-domain3.html
  2. https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf
  3. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/responsible-ai.html
  4. https://committee.iso.org/standard/42001
  5. https://docs.aws.amazon.com/aws-certification/latest/ai-business-strategist-01/ai-business-strategist-01-domain1.html
  6. https://aws.amazon.com/blogs/migration-and-modernization/govern-ai-adoption-before-it-governs-you/
  7. https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf
  8. https://oaic.gov.au/privacy/privacy-guidance-for-organisations-and-government-agencies/guidance-on-privacy-and-the-use-of-commercially-available-ai-products
  9. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/raibr03-bp01.html
  10. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/raibr03-bp02.html
  11. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/raibr03-bp03.html
  12. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/raibr03-bp04.html
  13. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/rairc01-bp01.html
  14. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/raiuc05-bp01.html
  15. https://aws.amazon.com/blogs/machine-learning/building-trust-in-ai-the-aws-approach-to-the-eu-ai-act/
  16. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/raier02-bp02.html
  17. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/raigt01-bp03.html
  18. https://docs.aws.amazon.com/bedrock/latest/userguide/data-protection.html
  19. https://docs.aws.amazon.com/IAM/latest/UserGuide/best-practices.html
  20. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/raisp03-bp02.html
  21. https://docs.aws.amazon.com/quick/latest/userguide/confluence-kb-acl.html
  22. https://docs.aws.amazon.com/wellarchitected/latest/security-pillar/sec_data_classification_define_protection.html
  23. https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html
  24. https://docs.aws.amazon.com/ai/responsible-ai/bedrock-guardrails/overview.html
  25. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/raimon01-bp03.html
  26. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/raimon01-bp01.html
  27. https://cheatsheetseries.owasp.org/cheatsheets/LLM_Prompt_Injection_Prevention_Cheat_Sheet.html
  28. https://docs.aws.amazon.com/wellarchitected/latest/responsible-ai-lens/raidp04-bp03.html
  29. https://committee.iso.org/standard/74438.html
  30. https://docs.aws.amazon.com/whitepapers/latest/aws-caf-for-ai/foundational-ai-capabilities.html