Ref: https://www.udemy.com/course/aws-ai-practitioner-certified/learn/lecture/44887843

• Contents:

Types of Security in AI

• Threat Detection
  • Example of threats: generating fake content, manipulated data, automated attacks
  • Deploy AI-based threat detection systems
    • Analyze NW traffic, user behavior, and other relevant data sources to identify threats
• Vulnerability Management
  • Identify vulnerabilities in AI systems: software bugs, model weaknesses...
  • Conduct security assessment, penetration testing, and code reviews
  • Patch management & update processes
• Infrastructure Protection
  • Secure the cloud computing platform, edge devices, and data stores
  • Access control, NW segmentation, encryption
  • Ensure you can withstand systems failures
• Data Encryption
  • Encrypt data at rest and in transit
  • Manage encryption keys properly and make sure they’re protected against unauthorized access
• Prompt Injection
  • Manipulated input prompts to generate malicious or undesirable content
  • Implement guardrails: prompt filtering, sanitization, validation
  • Additional reading: https://www.redsentry.com/blog/what-is-prompt-injection

Monitoring AI systems

• Performance Metrics
  • Model Accuracy – ratio of positive predictions
  • Precision – ratio of true positive predictions (correct vs. incorrect positive prediction)
  • Recall – ratio of true positive predictions compared to actual positive
  • F1-score – average of precision and recall (good balanced measure)
  • Latency – time taken by the model to make a prediction
• Infrastructure monitoring (catch bottlenecks and failures)
  • Compute resources (CPU and GPU usage)
  • NW performance
  • Storage
  • System Logs
• Bias and Fairness, Compliance and Responsible AI

AWS Shared Responsibility Model

• AWS responsibility → Security OF the Cloud
  • Protecting infrastructure (HW, SW, facilities, and NWing) that runs the AWS services
  • Managed services like Bedrock, SageMaker, S3…
• Customer responsibility → Security IN the Cloud
  • Encrypting application data
  • In Bedrock's case → customer responsible for data management, access controls, setting up guardrails, etc
• Shared controls:
  • Patch Management, Configuration Management, Awareness & Training
• Additional reading: https://aws.amazon.com/compliance/shared-responsibility-model/