Task 5.2: Recognize governance and compliance regulations for AI systems
Security, Compliance, and Governance for AI Solutions · 6,538 words · source: Certified-AI-Practitioner-AIF-C01/domain-5/task-5-2-cheat-sheet.md
Domain 5: Security, Compliance, and Governance for AI Solutions
Task Statement 5.2: Recognize governance and compliance regulations for AI systems
Domain 5 is "Security, Compliance, and Governance for AI Solutions" and represents 14% of the scored AIF-C01 exam content. Task 5.2 focuses on governance and compliance for AI systems: AWS services that support audits and regulation compliance, data governance strategies, logging, residency, monitoring, observability, retention, policies, review cadences, governance frameworks, transparency standards, and team training.
Use this sheet for questions such as "which AWS service helps prove compliance?", "how do you collect audit evidence?", "how should AI training data be governed?", "which service records API activity?", "how do you manage data retention?", "what changes when you fine-tune a model with company data?", and "how should an organization follow an AI governance protocol?"
Official Study Path
Use these in order:
- AWS Skill Builder: AWS Artificial Intelligence Practitioner Learning Plan
- AWS Skill Builder: Exam Prep Plan: AWS Certified AI Practitioner (AIF-C01)
- AWS Skill Builder: Domain 5 Review: AWS Certified AI Practitioner
- AWS Skill Builder: Domain 5 Practice: AWS Certified AI Practitioner
- AWS Exam Guide: Content Domain 5, Task Statement 5.2
- AWS Docs: AWS Config, Amazon Inspector, AWS Audit Manager, AWS Artifact, AWS CloudTrail, AWS Trusted Advisor, AWS Glue, AWS Glue Data Catalog, AWS Glue Data Quality, AWS Glue DataBrew, AWS Lake Formation, Amazon S3 lifecycle management, and AWS Generative AI Security Scoping Matrix
- Local Skill Builder transcripts: Task Statement 5.2, Lessons 1 through 6
Exam Objective Map
You should be able to answer questions about:
| Objective | What to know for the exam |
|---|---|
| AWS governance and compliance services | AWS Config, Amazon Inspector, AWS Audit Manager, AWS Artifact, AWS CloudTrail, AWS Trusted Advisor |
| Audit evidence and reports | Artifact provides AWS compliance reports; Audit Manager collects customer-side evidence and produces assessment reports |
| Data governance strategies | Data lifecycle, logging, residency, monitoring, observability, retention, curation, discovery, understanding, protection |
| Data governance roles | Data owner, data steward, IT, security, legal, compliance, model owner, business owner |
| AI standards and regulations | ISO/IEC 42001, ISO/IEC 23894, EU AI Act, NIST AI RMF, Algorithmic Accountability Act concepts |
| Risk management | Identify use case and stakeholders, estimate likelihood and impact, distinguish inherent and residual risk, document overall system risk |
| Governance protocols | Policies, review cadence, review strategies, transparency standards, employee training, governance frameworks |
| Generative AI scope | AWS Generative AI Security Scoping Matrix: responsibility increases as you move from consuming third-party apps to building and training your own models |
One-Minute Mental Model
Governance is the system for deciding who is accountable, what policies apply, how AI systems are reviewed, and how risks are monitored. Compliance is proof that the system follows required laws, standards, contracts, and internal policies.
Governance = people + policies + processes + controls + review.
Compliance = evidence that required controls are working.
Audit = independent review of that evidence.
Exam shortcut:
Artifact = AWS compliance reports.
Audit Manager = customer audit evidence and assessment reports.
CloudTrail = who did what, where, and when.
Config = resource configuration history and compliance rules.
Inspector = vulnerability findings for workloads and code.
Trusted Advisor = best-practice recommendations.
S3 lifecycle = retention, archival, deletion.
Lake Formation = fine-grained data lake governance.
Glue Data Catalog = metadata catalog.
Glue Data Quality = data quality rules and anomaly detection.
1. Governance vs Compliance vs Audit
These terms appear together, but they are not identical.
| Term | Exam meaning | AI example |
|---|---|---|
| Governance | How an organization directs and controls AI use | AI use policy, model approval board, data owner sign-off, review cadence |
| Compliance | Meeting external or internal requirements | SOC 2 controls, GDPR requirements, EU AI Act obligations, internal privacy policy |
| Audit | Review of controls and evidence | Auditor asks for access logs, data retention evidence, model risk documentation |
| Control | A safeguard or process used to reduce risk | Encryption, least privilege, access reviews, bias testing, output monitoring |
| Evidence | Proof that a control exists or operated | CloudTrail logs, Config compliance result, training record, policy document |
| Framework | Structured set of controls or practices | SOC 2, ISO 27001, ISO/IEC 42001, NIST AI RMF, AWS Audit Manager framework |
| Policy | Organization rule | "No customer PII may be used for fine-tuning without privacy approval" |
| Procedure | Repeatable steps to satisfy a policy | Review data classification, approve access, log sign-off, retain evidence |
Exam pattern:
Need proof for an auditor -> evidence and reports.
Need rules for AI use -> governance policy.
Need verify AWS service compliance posture -> AWS Artifact.
Need verify your AWS resource configuration -> AWS Config.
2. AWS Governance and Compliance Service Chooser
| Need | Best AWS answer | Why |
|---|---|---|
| Download AWS compliance reports and third-party attestations | AWS Artifact | Provides on-demand access to AWS and ISV security and compliance reports, certifications, and agreements |
| Collect evidence for customer audits | AWS Audit Manager | Maps frameworks to AWS data sources, collects evidence, and generates assessment reports |
| Use prebuilt compliance frameworks | AWS Audit Manager | Includes prebuilt frameworks for standards and regulations; can also use custom frameworks |
| Track resource configuration changes | AWS Config | Records configuration history and relationships for supported resources |
| Evaluate resources against compliance rules | AWS Config rules | Checks whether resources comply with desired configuration rules |
| Deploy compliance rules at scale | AWS Config conformance packs | Packages Config rules and remediation actions for accounts, Regions, or organizations |
| Automatically remediate a noncompliant resource configuration | AWS Config with Systems Manager Automation | Config can trigger remediation actions when rules fail |
| Record account activity and API calls | AWS CloudTrail | Records user activity and API calls as events |
| Answer "who did what, where, and when?" | AWS CloudTrail | Core audit trail service for governance, compliance, security monitoring, and troubleshooting |
| Store long-term API activity logs | CloudTrail trails to Amazon S3 | Trails deliver events to S3 and optionally CloudWatch Logs/EventBridge |
| Find software vulnerabilities and unintended network exposure | Amazon Inspector | Continuously scans supported workloads and code repositories and prioritizes findings |
| Get best-practice recommendations | AWS Trusted Advisor | Checks cost, performance, resilience, security, operational excellence, and service limits |
| Govern data lake access | AWS Lake Formation | Fine-grained permissions for data in S3 using AWS Glue Data Catalog metadata |
| Catalog data assets and schemas | AWS Glue Data Catalog | Metadata store for data locations, schemas, data types, and table definitions |
| Check data quality rules | AWS Glue Data Quality | Evaluates data quality rules and can use ML to detect anomalies |
| Profile, clean, and prepare data visually | AWS Glue DataBrew | Data profiling, cleaning, normalization, and lineage-style preparation views |
| Manage retention, archival, and deletion | Amazon S3 lifecycle rules | Transitions or expires objects based on time and storage class policies |
| Discover sensitive data in S3 | Amazon Macie | Finds sensitive data such as PII; useful before training, fine-tuning, or RAG |
| Filter GenAI inputs and outputs | Amazon Bedrock Guardrails | Supports content filters, denied topics, PII handling, and grounding checks |
3. Shared Responsibility for Compliance
AWS Skill Builder emphasizes that the shared responsibility model applies to both security and compliance.
| Responsibility area | AWS responsibility | Customer responsibility |
|---|---|---|
| Physical infrastructure | Secure data centers, hardware, networking, and facilities | Use AWS compliance reports as inherited evidence |
| Managed services | Operate and secure the service infrastructure | Configure service access, logging, encryption, data use, and retention |
| Compliance reports | Provide third-party audit reports and attestations for AWS controls | Determine which standards apply to the workload and provide customer-side evidence |
| AI data | Provide security and governance features | Classify, protect, retain, delete, and monitor data used for AI |
| AI model and app behavior | Provide service controls and documentation | Choose models, document risks, add guardrails, monitor outputs, train teams, review policies |
Exam Pattern
If the question asks who provides the SOC 2 report for AWS data center controls, answer AWS through AWS Artifact.
If the question asks who proves that the application team configured encryption, access controls, logging, and retention correctly, answer the customer, often using AWS Audit Manager, AWS Config, CloudTrail, and related evidence.
If the question asks why AWS reports reduce audit scope, the answer is inherited controls: AWS has already had its side audited by third parties, so the customer's auditors can focus on the customer's workload and processes.
4. AWS Artifact
AWS Artifact is the exam answer when the organization needs AWS compliance documentation.
Use AWS Artifact when the question says:
- Download AWS SOC, ISO, PCI, or other compliance reports.
- Provide AWS third-party auditor reports to internal or external auditors.
- Understand the compliance and security posture of AWS.
- Review, accept, or manage select AWS agreements.
- Assess security and compliance reports for some ISVs on AWS Marketplace.
What AWS Artifact is not:
| Not this | Use instead |
|---|---|
| Evidence that your workload is configured correctly | AWS Audit Manager, AWS Config, CloudTrail |
| Vulnerability scanner | Amazon Inspector |
| Runtime monitoring service | Amazon CloudWatch, CloudTrail, Security Hub, GuardDuty depending on scenario |
| Data catalog | AWS Glue Data Catalog |
Memory hook:
Artifact = AWS gives you AWS-side audit artifacts.
5. AWS Audit Manager
AWS Audit Manager helps continuously audit AWS usage and simplify risk and compliance assessment.
What Audit Manager Does
| Capability | Exam meaning |
|---|---|
| Prebuilt frameworks | Frameworks map controls to AWS data sources for standards such as SOC 2, PCI DSS, HIPAA, GDPR, NIST, and others |
| Custom frameworks | Create controls and frameworks for internal policies or special audit needs |
| Automated evidence collection | Collects evidence from AWS services such as CloudTrail, AWS Config, Security Hub, and License Manager |
| Multi-account evidence collection | Uses AWS Organizations to collect and consolidate evidence across accounts |
| Delegation workflow | Assigns control sets to responsible team members for review |
| Evidence finder | Helps search and export assessment evidence |
| Assessment report | Auditor-ready package of selected evidence and assessment summary |
Audit Manager Exam Pattern
Choose AWS Audit Manager when the question says:
- "Automate evidence collection for an audit."
- "Map compliance requirements to AWS usage data."
- "Generate an assessment report for auditors."
- "Use a framework to assess controls."
- "Collect evidence across multiple AWS accounts."
- "Demonstrate that controls are operating as intended."
Do not confuse:
Audit Manager = collects and organizes audit evidence.
Artifact = provides AWS compliance reports.
Config = evaluates resource configuration compliance.
CloudTrail = records API activity used as evidence.
6. AWS Config
AWS Config is the exam answer for resource configuration history, resource relationships, and compliance evaluation against configuration rules.
What Config Does
| Capability | Exam meaning |
|---|---|
| Configuration history | Records what a resource configuration looked like over time |
| Resource relationships | Tracks relationships between resources, such as security groups and EC2 instances |
| Config rules | Evaluates resources against built-in or custom rules |
| Custom rules | Use Lambda or policy-as-code logic for organization-specific checks |
| Conformance packs | Package rules and remediation actions into a deployable compliance baseline |
| Aggregators | Centralize configuration and compliance data across accounts and Regions |
| Remediation | Can invoke Systems Manager Automation documents to fix noncompliant resources |
AI and ML Compliance Examples
| Requirement | Config pattern |
|---|---|
| S3 buckets containing training data must block public access | Config rule for S3 public access settings |
| CloudTrail must be enabled | Config rule or conformance pack |
| SageMaker notebooks must not allow direct internet access | Config rule or custom rule |
| Encryption must be enabled for storage resources | Config managed rule or custom rule |
| AI/ML baseline must be applied across accounts | Conformance pack such as operational best practices for AI/ML or SageMaker security best practices |
Exam shortcut:
Configuration drift or noncompliant resource setting -> AWS Config.
Compliance evidence report -> Audit Manager.
API activity trail -> CloudTrail.
7. AWS CloudTrail
AWS CloudTrail supports auditing, security monitoring, operational troubleshooting, and compliance by recording user activity and API calls across AWS services.
CloudTrail Event Types
| Event type | What it captures | Example |
|---|---|---|
| Management events | Control plane actions | Create bucket, update IAM policy, invoke Bedrock model configuration API |
| Data events | Data plane actions for supported resources | Read or write S3 object, invoke Lambda |
| Network activity events | Actions through VPC endpoints, including denied API calls | Private API access attempts |
| Insights events | Unusual API activity or error rates | Spike in failed API calls |
CloudTrail Exam Pattern
Choose CloudTrail when the question says:
- "Who changed this configuration?"
- "Who accessed this API?"
- "What action happened, from which identity, source IP, and time?"
- "Provide an audit trail for AI interactions with AWS services."
- "Store API activity logs in S3 for long-term retention."
- "Monitor or alert on account activity."
CloudTrail supports audit trails, but be careful:
CloudTrail logs AWS API activity.
Application-level AI chat logs are usually logged by the application to CloudWatch Logs, S3, or another logging system.
Current-doc note:
As of the AWS CloudTrail features page, CloudTrail Lake is no longer open to new customers starting May 31, 2026. For AIF-C01, focus on CloudTrail Event History and trails unless the question specifically asks about querying event data.
8. Amazon Inspector
Amazon Inspector is a vulnerability management service. It continually scans supported workloads and code repositories for software vulnerabilities and unintended network exposure.
What Inspector Scans
| Target | Exam meaning |
|---|---|
| Amazon EC2 instances | Software vulnerabilities and unintended network exposure |
| Amazon ECR container images | Vulnerabilities in container images |
| AWS Lambda functions | Package vulnerabilities and code-related risks |
| Code repositories | Code security findings where supported |
Inspector Exam Pattern
Choose Amazon Inspector when the question says:
- "Find vulnerabilities in workloads."
- "Assess containers for CVEs."
- "Prioritize vulnerability findings by severity or risk."
- "Detect unintended network exposure."
- "Continuously scan compute workloads."
- "Route vulnerability findings to Security Hub or EventBridge."
Do not choose Inspector for:
| Need | Better answer |
|---|---|
| Prove compliance with an auditor report | Audit Manager |
| View AWS compliance certificates | Artifact |
| Record API activity | CloudTrail |
| Evaluate resource configuration rules | Config |
| Find sensitive data in S3 | Macie |
9. AWS Trusted Advisor
AWS Trusted Advisor continuously evaluates an AWS environment against best-practice checks and recommends actions.
| Category | What it helps with |
|---|---|
| Cost optimization | Unused or underused resources, cost-saving opportunities |
| Performance | Configuration and usage recommendations |
| Resilience | Redundancy and availability checks |
| Security | Security gaps and best-practice deviations |
| Operational excellence | Operational recommendations |
| Service limits | Usage approaching or exceeding quotas |
Exam pattern:
Best-practice recommendations across cost, performance, resilience, security, operations, and service limits -> Trusted Advisor.
Trusted Advisor is advisory. It recommends actions. It is not the main source of formal audit evidence, configuration history, or compliance reports.
10. AI Standards and Regulations
The exam does not require you to be a lawyer. It expects you to recognize why governance, risk management, transparency, and compliance evidence matter.
| Standard or regulation | Exam meaning |
|---|---|
| ISO/IEC 42001:2023 | AI management system standard for establishing, implementing, maintaining, and improving AI governance and management |
| ISO/IEC 23894:2023 | Guidance for managing risk related to AI systems and integrating AI risk management into organizational activities |
| EU AI Act | Risk-based AI regulation; bans unacceptable-risk systems and creates obligations for high-risk and certain transparency-sensitive systems |
| NIST AI Risk Management Framework | Voluntary framework for managing AI risks through Govern, Map, Measure, and Manage functions |
| Algorithmic Accountability Act concept | Proposed US legislation concept focused on impact assessments, transparency, and protection from unfair or unexplained automated decisions |
| GDPR / CCPA concepts | Privacy laws that influence data governance, consent, retention, deletion, and personal data handling |
| SOC 2 | Assurance report covering controls related to security, availability, processing integrity, confidentiality, and privacy |
| ISO 27001 | Information security management standard; relevant to security governance and control management |
EU AI Act Risk Pattern
For exam study, know the risk-based idea:
| Risk level | Practical meaning |
|---|---|
| Unacceptable risk | Prohibited or banned because the use is incompatible with rights or safety |
| High risk | Subject to stricter requirements such as risk management, data governance, documentation, logging, transparency, human oversight, and accuracy/security controls |
| Limited risk | Transparency obligations, such as telling users they are interacting with AI in certain cases |
| Minimal or no risk | Lighter obligations |
Skill Builder examples of prohibited or high-risk themes:
- Social scoring.
- Scraping facial images to create facial recognition databases.
- Emotion inference in workplaces or educational institutions.
- Employment screening tools that rank applicants.
Exam shortcut:
Regulated or high-impact AI decision -> risk assessment + documentation + transparency + monitoring + human oversight + compliance evidence.
11. NIST AI RMF and Risk Management
AWS Skill Builder emphasizes risk as a combination of likelihood and severity.
Risk = likelihood of event x severity of consequence.
Risk Assessment Flow
- Identify the AI use case.
- Identify stakeholders affected by the system.
- Identify potentially harmful events.
- Estimate likelihood and severity for each event.
- Determine inherent risk before controls.
- Apply controls such as access control, monitoring, guardrails, testing, human review, or data governance.
- Determine residual risk after controls.
- Summarize the system's overall risk by focusing on the highest residual risks.
- Review and update the assessment on a defined cadence.
NIST AI RMF Functions
| Function | Exam meaning |
|---|---|
| Govern | Establish policies, accountability, roles, culture, and risk management structure |
| Map | Contextualize the AI system, use case, stakeholders, benefits, and risks |
| Measure | Analyze, assess, benchmark, test, and monitor AI risks and trustworthiness |
| Manage | Prioritize, respond to, mitigate, monitor, and communicate risks |
Memory hook:
Govern = set rules.
Map = understand context.
Measure = assess risk.
Manage = act on risk.
12. Explainability, Transparency, and Bias in Compliance
Task 5.2 overlaps with Domain 4 because compliance questions often ask whether an AI system can be explained, audited, and monitored for bias.
| Requirement | What it means | AWS service or practice |
|---|---|---|
| Disclose AI use | Tell users when they interact with AI where required or appropriate | UX disclosure, chatbot notice, transparency policy |
| Explain output | Provide understandable reasons for model predictions or decisions | SageMaker Clarify, model cards, documentation |
| Test bias | Check whether model outputs unfairly affect groups | SageMaker Clarify, subgroup analysis, human review |
| Monitor drift | Watch whether bias or feature attribution changes over time | SageMaker Clarify, SageMaker Model Monitor |
| Document intended use | Record what the model is for and not for | SageMaker Model Cards, AI service cards, internal documentation |
| Add human review | Escalate uncertain or high-impact decisions | Amazon A2I, business review process |
Exam Pattern
If the question asks how to understand which variables influence a model's behavior or monitor for bias and feature attribution drift, choose Amazon SageMaker Clarify.
If the question asks how to document model purpose, risk, intended use, and evaluation, choose Amazon SageMaker Model Cards.
If the question asks how to meet transparency expectations for a chatbot, choose disclosure, documentation, logging, and clear user-facing boundaries.
13. Data Governance Strategy
AWS Skill Builder defines data governance as the combination of people, process, and technology used to manage the availability, usability, integrity, and security of enterprise data.
Three Major Parts
| Part | Meaning | AI example |
|---|---|---|
| Curation | Identify, manage, clean, and maintain valuable data sources | Curate approved training or RAG documents |
| Discovery and understanding | Make data findable and understandable through metadata and context | Use a data catalog with owners, schemas, sensitivity, and lineage |
| Protection | Balance privacy, security, and access | Apply Lake Formation permissions, encryption, Macie scans, retention rules |
Data Governance Roles
| Role | Responsibility |
|---|---|
| Data owner | Executive or accountable business owner who makes policy decisions for a data domain |
| Data steward | Business expert who understands the data and handles day-to-day data quality and usage details |
| IT / platform team | Provides systems, tools, catalogs, access controls, monitoring, and automation |
| Security team | Defines security controls, monitors threats, and validates control operation |
| Legal / compliance team | Interprets laws, contracts, and regulatory obligations |
| Model owner | Owns model risk, intended use, evaluation, and lifecycle decisions |
| Business owner | Owns business outcome, acceptable risk, and user impact |
Data Governance Vocabulary
| Term | Exam meaning |
|---|---|
| Data profiling | Systematically examining data characteristics and detecting quality issues |
| Data quality | Accuracy, completeness, consistency, timeliness, and fitness for use |
| Data catalog | Metadata inventory of data sources, schemas, types, owners, and definitions |
| Data lineage | Origin, movement, transformations, storage, and downstream use of data |
| Data lifecycle | Creation or ingestion through storage, use, archive, retention, deletion |
| Data residency | Requirement that data is stored or processed in specific geographic locations |
| Data retention | How long data must be kept before archival or deletion |
| Observability | Ability to understand system state from logs, metrics, traces, events, and monitoring |
| Monitoring | Continuous checking of systems, models, data quality, compliance, drift, and bias |
14. AWS Data Governance Services
| Service | Governance role | Exam clue |
|---|---|---|
| AWS Glue Data Catalog | Stores metadata such as data locations, schemas, data types, and table definitions | "Catalog data sources" |
| AWS Glue crawlers | Scan data sources and populate the Data Catalog | "Automatically discover schemas" |
| AWS Glue Data Quality | Evaluates data quality rules for objects in the Data Catalog | "Recommend or run data quality rules" |
| AWS Glue DataBrew | Visual data preparation, cleaning, normalization, profiling | "Clean and normalize without writing code" |
| AWS Lake Formation | Fine-grained governance for data lakes in S3 using the Glue Data Catalog | "Column, row, or cell-level access control" |
| Amazon S3 lifecycle | Transition, archive, or delete objects based on lifecycle rules | "Retain training data for compliance at lower cost" |
| Amazon S3 storage classes | Match storage cost to access pattern | "Archive rarely accessed training data" |
| Amazon Macie | Sensitive data discovery in S3 | "Find PII before training or RAG" |
| AWS CloudTrail | API activity logs | "Audit who accessed or changed resources" |
| Amazon CloudWatch | Metrics, logs, alarms, dashboards | "Operational monitoring and observability" |
Data Governance Service Patterns
| Scenario | Best answer |
|---|---|
| Data consumers need to find approved datasets | AWS Glue Data Catalog or data cataloging strategy |
| Need to restrict analysts to specific rows and columns in a data lake | AWS Lake Formation |
| Need to assess quality of a dataset before using it for model training | AWS Glue Data Quality or DataBrew profiling |
| Need to clean and normalize a dataset visually | AWS Glue DataBrew |
| Need to retain training data for 5 years and archive it cheaply | S3 lifecycle rules to Glacier storage classes |
| Need to delete data after a retention period | S3 lifecycle expiration rule |
| Need to know where a report field came from | Data lineage |
| Need to know whether a dataset contains PII | Amazon Macie or sensitive data detection |
15. Data Lifecycle, Retention, and Residency
AI governance should treat data as a lifecycle asset, not as a one-time input.
| Stage | Governance questions |
|---|---|
| Create or collect | What is the source? Is collection lawful? Is consent or notice required? |
| Classify | Is the data public, internal, confidential, regulated, or personal? |
| Store | Which Region? Which account? Which bucket? Which encryption key? |
| Catalog | Who owns it? What schema? What business definition? What sensitivity? |
| Access | Which roles can use it? Is access temporary? Is approval required? |
| Process | What transformations occurred? Are quality rules passing? |
| Train, tune, or retrieve | Is the data approved for AI use? Can it be used in prompts, RAG, or fine-tuning? |
| Monitor | Are there access logs, quality checks, drift checks, and compliance checks? |
| Retain | How long must it be kept for audit, legal, or business needs? |
| Archive | Which S3 storage class fits access and retrieval requirements? |
| Delete | When must it be expired? Are deletion requirements documented? |
S3 Storage Class Exam Pattern
| Access pattern | Likely storage class |
|---|---|
| Frequently accessed training or inference data | S3 Standard |
| Infrequently accessed but needs quick retrieval | S3 Standard-IA or S3 One Zone-IA |
| Unknown or changing access patterns | S3 Intelligent-Tiering |
| Rarely accessed archive for retention or compliance | S3 Glacier storage classes |
| Very long-term archive with lowest storage cost and slower retrieval | S3 Glacier Deep Archive |
Exam shortcut:
Known retention timeline -> S3 lifecycle rule.
Unknown access pattern -> S3 Intelligent-Tiering.
Long-term compliance archive -> S3 Glacier storage classes.
16. Generative AI Security Scoping Matrix
The AWS Generative AI Security Scoping Matrix helps organizations understand responsibility based on how generative AI is consumed or built.
| Scope | Pattern | Responsibility level |
|---|---|---|
| Scope 1 | Consumer app | Lowest: organization consumes a public third-party AI application |
| Scope 2 | Enterprise app | Low: organization consumes a managed enterprise AI application with administrative controls |
| Scope 3 | Pre-trained model application | Higher: organization builds an app using a pre-trained model |
| Scope 4 | Fine-tuned model application | Higher still: organization adapts a model with its own data |
| Scope 5 | Self-trained model | Highest: organization builds and trains its own model |
Skill Builder exam takeaway:
Minimize scope when possible.
More ownership of model and data = more responsibility for governance, compliance, privacy, risk management, security controls, and resilience.
Left-to-Right Selection Pattern
When solving a business problem, Skill Builder recommends starting with the lowest-responsibility option that meets the requirement:
- Use a fully trained AWS AI service if it fits, such as Amazon Comprehend or Amazon Translate.
- Use a pre-trained foundation model through Amazon Bedrock if the use case needs generative AI.
- Add RAG through Amazon Bedrock Knowledge Bases if the model needs private knowledge.
- Fine-tune or customize a model only when lower-scope approaches do not meet requirements.
- Train a custom model only when necessary.
This reduces governance burden, legal/privacy risk, operational complexity, security-control ownership, and model-resilience responsibility.
17. Governance Protocol Process
A practical AI governance process looks like this:
- Identify scope of responsibility using the Generative AI Security Scoping Matrix.
- Define the use case, users, stakeholders, and business owner.
- Classify data and model risk.
- Identify applicable regulations, standards, contracts, and internal policies.
- Define policies for data governance, access, model transparency, logging, retention, human review, and incident response.
- Assign roles such as data owner, data steward, model owner, security owner, compliance reviewer, and business approver.
- Select controls and AWS services to enforce and monitor policies.
- Document the system, risks, controls, model purpose, and intended use.
- Train employees according to job role and access level.
- Monitor performance, compliance, security, data quality, bias, and drift.
- Review results on a defined cadence.
- Update policies, controls, and training as risks or business goals change.
Review Cadence Examples
| Trigger | Review needed |
|---|---|
| New model, dataset, tool, or agent capability | Architecture, security, legal, and model-risk review |
| New regulation or contract requirement | Compliance gap analysis |
| Failed Config rule or Inspector critical finding | Remediation review |
| Increased hallucination, toxicity, bias, or drift | Responsible AI and model-performance review |
| New data domain used for RAG or fine-tuning | Data owner and privacy review |
| User complaint or incident | Incident response and governance review |
| Scheduled quarterly or annual review | Control effectiveness and policy refresh |
18. Transparency Standards and Team Training
Governance protocols only work when people understand their responsibilities.
Transparency Standards
| Standard | What to document |
|---|---|
| User transparency | When users interact with AI, limitations, escalation paths, and human review options |
| Model transparency | Model purpose, intended use, limitations, risk rating, evaluation, monitoring, and known constraints |
| Data transparency | Data sources, owners, lineage, sensitivity, allowed uses, and retention |
| Decision transparency | How AI influences decisions, what humans review, and how decisions can be challenged |
| Control transparency | Which controls exist, how they are monitored, and what evidence is retained |
Training Requirements
| Team | Training focus |
|---|---|
| Business users | Approved AI use, data handling, limitations, disclosure rules |
| Developers | Secure AI design, logging, guardrails, least privilege, prompt injection risks |
| Data stewards | Data quality, cataloging, lineage, sensitivity, retention |
| Model owners | Evaluation, bias, drift, transparency, model cards, review cadence |
| Security team | Threat modeling, logging, vulnerability management, incident response |
| Legal/compliance | Regulations, evidence, audit readiness, data residency, privacy obligations |
| Support teams | User escalation, incident reporting, AI output limitations |
Exam shortcut:
Governance is not only tooling. It also includes policies, owners, reviews, and training.
19. Bedrock Guardrails in Governance Context
Amazon Bedrock Guardrails appears in Task 5.1, but it can also show up in governance scenarios because it implements application-specific responsible AI policies.
| Governance need | Guardrails feature |
|---|---|
| Block harmful categories | Content filters |
| Avoid restricted business topics | Denied topics |
| Detect or redact PII | Sensitive information filters |
| Reduce unsafe prompt behavior | Prompt attack filters |
| Check support from source context | Contextual grounding checks |
| Provide consistent blocked-message behavior | Custom blocked input and response messages |
Exam pattern:
Policy says chatbot must not provide investment advice -> Bedrock Guardrails denied topic.
Policy says chatbot must not expose PII -> Guardrails sensitive information filter plus data governance controls.
Policy says responses must be grounded in retrieved sources -> RAG plus grounding checks and citations.
20. Common Exam Traps
| Trap | Correct thinking |
|---|---|
| "Compliance is fully handled by AWS" | False. AWS handles compliance of the cloud; customers handle compliance in the cloud |
| "Artifact proves my workload is compliant" | Artifact provides AWS reports. Your workload compliance needs your evidence |
| "CloudTrail evaluates compliance rules" | CloudTrail records API activity. Config evaluates resource configuration compliance |
| "Inspector checks data quality" | Inspector checks vulnerabilities and exposure. Glue Data Quality checks data quality |
| "Trusted Advisor creates audit reports" | Trusted Advisor recommends best-practice actions. Audit Manager creates assessment reports |
| "Data retention means keep everything forever" | Retention includes keep, archive, and delete according to policy and law |
| "Fine-tuning is only a model-quality decision" | Fine-tuning with company data increases data governance, privacy, compliance, and model-risk responsibility |
| "A model can enforce data permissions by itself" | Access control must be enforced before retrieval and prompt construction |
| "Transparency only means publishing model code" | Transparency can include disclosure, documentation, explanations, logs, model cards, and review paths |
| "Governance frameworks replace controls" | Frameworks organize controls; teams still implement and verify them |
21. Scenario Decision Table
| Scenario question says... | Choose... |
|---|---|
| "The auditor asks for AWS SOC 2 report" | AWS Artifact |
| "The auditor asks for evidence that your S3 training buckets are encrypted" | AWS Audit Manager with evidence from AWS Config |
| "The security team needs to know who changed a model endpoint configuration" | AWS CloudTrail |
| "The compliance team needs to know whether resources drifted from policy" | AWS Config |
| "The platform team wants a reusable compliance baseline across accounts" | AWS Config conformance packs |
| "The application team needs to find critical CVEs in container images" | Amazon Inspector |
| "The operations team wants recommended actions for security and service limits" | AWS Trusted Advisor |
| "A data owner needs to approve which datasets can be used for fine-tuning" | Data governance policy and data owner approval |
| "A business user needs to find trusted customer datasets" | AWS Glue Data Catalog or data catalog |
| "A team must prevent analysts from seeing sensitive columns" | AWS Lake Formation fine-grained access control |
| "Training data must be retained for five years at low cost" | S3 lifecycle rules to Glacier storage classes |
| "A chatbot must disclose it is AI" | Transparency standard and user-facing disclosure |
| "A high-impact model must be reviewed for bias" | SageMaker Clarify, human review, governance process |
| "The team is choosing between fully managed AI service and custom model" | Use Generative AI Security Scoping Matrix and minimize scope |
22. Mini Flashcards
| Question | Answer |
|---|---|
| What is Domain 5 worth? | 14% of scored AIF-C01 content |
| What is the main Task 5.2 theme? | Governance and compliance regulations for AI systems |
| What AWS service provides AWS compliance reports? | AWS Artifact |
| What AWS service collects audit evidence and creates assessment reports? | AWS Audit Manager |
| What AWS service records API calls and user activity? | AWS CloudTrail |
| What AWS service tracks resource configuration history and compliance? | AWS Config |
| What AWS feature packages Config rules and remediation actions? | Conformance packs |
| What AWS service scans for vulnerabilities and unintended network exposure? | Amazon Inspector |
| What AWS service gives best-practice recommendations across cost, performance, resilience, security, operations, and limits? | AWS Trusted Advisor |
| What AWS service governs fine-grained data lake permissions? | AWS Lake Formation |
| What stores metadata about data locations, schemas, and table definitions? | AWS Glue Data Catalog |
| What evaluates data quality rules in Glue? | AWS Glue Data Quality |
| What service helps clean and normalize data visually? | AWS Glue DataBrew |
| What manages S3 archival and deletion over time? | S3 lifecycle rules |
| What is data lineage? | The origin, movement, transformation, storage, and downstream use of data |
| What is residual risk? | Risk remaining after controls and mitigations |
| What are the NIST AI RMF functions? | Govern, Map, Measure, Manage |
| What happens as GenAI scope increases? | Customer responsibility for governance, compliance, security, privacy, and resilience increases |
23. Last-Minute Checklist
Before the exam, make sure you can say yes to each statement:
- I know Domain 5 is 14% of the scored exam.
- I know Task 5.2 is about governance, compliance, data governance, and governance protocols.
- I can distinguish AWS Artifact, Audit Manager, Config, CloudTrail, Inspector, and Trusted Advisor.
- I know Artifact is AWS-side reports, while Audit Manager helps collect customer-side evidence.
- I know CloudTrail records API activity, while Config evaluates resource configurations.
- I know Inspector is for vulnerability management, not audit reports.
- I know Trusted Advisor provides best-practice recommendations.
- I know data governance includes curation, discovery and understanding, and protection.
- I know data governance includes lifecycle, logging, residency, monitoring, observability, and retention.
- I can explain data owner vs data steward.
- I can identify Glue Data Catalog, Glue Data Quality, DataBrew, Lake Formation, and S3 lifecycle patterns.
- I can explain inherent risk vs residual risk.
- I know the NIST AI RMF functions: Govern, Map, Measure, Manage.
- I understand that EU AI Act questions are risk-based: unacceptable risk is prohibited; high-risk systems have stronger obligations.
- I know governance protocols require policies, owners, review cadences, transparency standards, and training.
- I know the Generative AI Security Scoping Matrix: more model/data ownership means more responsibility.
Sources
- AWS Certified AI Practitioner Exam Guide, Domain 5: https://docs.aws.amazon.com/aws-certification/latest/ai-practitioner-01/ai-practitioner-01-domain5.html
- AWS Skill Builder: AWS Artificial Intelligence Practitioner Learning Plan: https://explore.skillbuilder.aws/learn/learning-plans/2217/aws-artificial-intelligence-practitioner-learning-plan
- AWS Skill Builder: Exam Prep Plan: AWS Certified AI Practitioner (AIF-C01): https://skillbuilder.aws/learning-plan/3NRN71QZR2/exam-prep-plan-aws-certified-ai-practitioner-aifc01--english/FBV4STG94B
- AWS Skill Builder: Domain 5 Review: AWS Certified AI Practitioner: https://skillbuilder.aws/learn/9Q5Y91H453/domain-5-review-aws-certified-ai-practitioner--aifc01--english/VQUGJZ6W7K
- AWS Skill Builder: Domain 5 Practice: AWS Certified AI Practitioner: https://skillbuilder.aws/learn/QZ8JXA94RS/domain-5-practice-aws-certified-ai-practitioner--aifc01--english/26PPD2C8KV
- Local Skill Builder transcript references:
.agent/skills/aws-builder-content/references/Domain-5/task-statement-5-2-lesson-1.mdthroughlesson-6.md - Local learning path reference:
.agent/skills/aws-learning-path/references/Domain-5-Security-Compliance-and-Fvernance-for-AI-Solutions.md - AWS Artifact: https://aws.amazon.com/artifact/
- AWS Audit Manager features: https://aws.amazon.com/audit-manager/features/
- AWS Config features: https://aws.amazon.com/config/features/
- AWS CloudTrail features: https://aws.amazon.com/cloudtrail/features/
- Amazon Inspector features: https://aws.amazon.com/inspector/features/
- AWS Trusted Advisor: https://aws.amazon.com/premiumsupport/technology/trusted-advisor/
- AWS data governance whitepaper: https://docs.aws.amazon.com/whitepapers/latest/enterprise-data-governance-catalog/data-governance.html
- AWS Lake Formation features: https://aws.amazon.com/lake-formation/features/
- Amazon S3 lifecycle management: https://docs.aws.amazon.com/AmazonS3/latest/userguide/object-lifecycle-mgmt.html
- AWS Glue Data Quality: https://docs.aws.amazon.com/glue/latest/dg/data-quality-gs-studio.html
- AWS Generative AI Security Scoping Matrix: https://aws.amazon.com/ai/security/generative-ai-scoping-matrix/
- AWS Security Blog: Securing generative AI, data, compliance, and privacy considerations: https://aws.amazon.com/blogs/security/securing-generative-ai-data-compliance-and-privacy-considerations/
- European Commission: AI Act: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
- NIST AI Risk Management Framework: https://www.nist.gov/itl/ai-risk-management-framework
- NIST AI RMF Core: https://airc.nist.gov/airmf-resources/airmf/5-sec-core/
- ISO/IEC 42001:2023: https://www.iso.org/standard/42001
- ISO/IEC 23894:2023: https://www.iso.org/standard/77304.html