Domain 5: Security, Compliance, and Governance (14%)
5.1: Securing AI Systems
Data Security
AI Security
1 / 3
❓
What is Prompt Injection?
(Click to reveal)💡
Malicious prompts to bypass safety controls
Example: "Ignore previous instructions and..."
Protection: Input sanitization, rate limiting, monitoring.
Example: "Ignore previous instructions and..."
Protection: Input sanitization, rate limiting, monitoring.
Encryption:
- ✅ At Rest: AWS KMS, S3 encryption
- ✅ In Transit: TLS/SSL, HTTPS
Access Control:
- IAM policies for fine-grained permissions
- VPC for network isolation
- Security groups and NACLs
Input Validation:
- Sanitize user inputs
- Limit input size
- Check for injection attacks
Output Filtering:
- Content moderation
- PII detection and redaction
- Toxicity filtering
Model Security
Adversarial Attacks:
- Prompt Injection: Malicious prompts to bypass safety
- Data Poisoning: Corrupt training data
- Model Inversion: Extract training data
Protection:
- Input sanitization
- Rate limiting
- Monitoring for unusual patterns
- Regular security audits
5.2: Governance and Compliance
Model Governance
Version Control:
- SageMaker Model Registry
- Track model lineage
- Associate models with data/code
Approval Workflows:
- Manual approval before production
- Automated testing gates
- Change management
Audit Trails:
- CloudTrail logs all API calls
- SageMaker logs training jobs
- Model deployment history
Compliance
Compliance Requirements
1 / 3
❓
What is HIPAA compliance for AI?
(Click to reveal)💡
Healthcare data protection
Use HIPAA-eligible AWS services
Encrypt PHI, access controls, audit logs.
Use HIPAA-eligible AWS services
Encrypt PHI, access controls, audit logs.
Data Residency:
- Choose appropriate AWS Region
- Keep data in specific geographic location
Industry Regulations:
- HIPAA: Healthcare data (use HIPAA-eligible services)
- GDPR: EU data privacy (right to explanation)
- SOC 2: Security controls
- PCI DSS: Payment card data
Documentation Requirements:
- Model cards
- Training data sources
- Performance metrics
- Bias testing results
Quick Reference: AWS AI/ML Services
When to Use Which Service?
| Need | Service |
|---|---|
| Access foundation models | Amazon Bedrock |
| Business assistant | Amazon Q |
| Code suggestions | Amazon CodeWhisperer |
| End-to-end ML platform | Amazon SageMaker |
| Image/video analysis | Amazon Rekognition |
| Extract text from documents | Amazon Textract |
| Sentiment analysis | Amazon Comprehend |
| Translation | Amazon Translate |
| Speech-to-text | Amazon Transcribe |
| Text-to-speech | Amazon Polly |
| Chatbot | Amazon Lex |
| Recommendations | Amazon Personalize |
| Fraud detection | Amazon Fraud Detector |
| Intelligent search | Amazon Kendra |
| Human review | Amazon Augmented AI (A2I) |
| Vector database for RAG | Amazon OpenSearch Service |
| Detect bias | SageMaker Clarify |
| Monitor models | SageMaker Model Monitor |