Vertex AI and Generative AI Security

AI practitioners, security professionals, and cloud architects

After this course participants should be able to:

• Review Google Cloud Security fundamentals.
• Establish a foundational understanding of Vertex AI.
• Enumerate the security concerns related to Vertex AI features and components.
• Control access with Identity Access Management.
• Simplify permission using organization hierarchies and policies. 
• Use service accounts for least privileged access.
• Configure encryption at rest and in-transit.
• Encrypt data using customer-managed encryption keys.
• Protect sensitive data using the Data Loss Prevention service.
• Prevent exfiltration of data using VPC Service Controls.
• Architect systems with disaster recovery in mind.
• Deploy ML models using model endpoints.
• Secure model endpoints.
• Write to and analyze logs.
• Set up monitoring and alerting.
• Identify security risks specific to LLMs and gen AI applications.
• Understand methods for mitigating prompt hacking and injection attacks.
• Explore the fundamentals of securing generative AI models and applications.
• Introduce fundamentals of AI Safety.
• Implement best practices for testing model responses.
• Apply techniques for improving response security in gen AI applications.
• Understand RAG architecture and security implications.
• Implement best practices for grounding and securing data sources in RAG systems.

Module 1: Introduction to Vertex AI Security Principles

• Google Cloud Security
• Vertex AI components
• Vertex AI Security concerns
• Lab: Vertex AI: Training and Serving a Custom Model

Module 2: Identity and Access Management (IAM) in Vertex AI

• Overview of IAM in Google Cloud
• Lab: Service Accounts and Roles: Fundamentals

Module 3: Data Security and Privacy

• Data encryption
• Protecting Sensitive Data
• VPC Service Controls
• Disaster recovery planning
• Lab: Getting Started with Cloud KMS
• Lab: Creating a De-identified Copy of Data in Cloud Storage

Module 4: Securing Vertex AI Endpoints and Model Deployment

• Network security
• Securing model endpoints
• Lab: Configuring Private Google Access and Cloud NAT

Module 5: Monitoring and Logging in Vertex AI

• Logging
• Monitoring

Module 6: Security Risks in Generative AI Applications

• Overview of gen AI security risks
• Overview of AI Safety
• Prompt security
• LLM safeguards
• Lab: Safeguarding with Vertex AI Gemini API
• Lab: Gen AI & LLM Security for Developers

Module 7: Testing and Evaluating Generative AI Model Responses

• Testing generative AI model responses
• Evaluating model responses
• Fine-Tuning LLMs
• Lab: Measure Gen AI Performance with the Generative AI Evaluation Service
• Lab: Unit Testing Generative AI Applications

Module 8: Securing Retrieval-Augmented Generation (RAG) Systems

• Fundamentals of Retrieval-Augmented Generation
• Security in RAG systems
• Lab: Multimodal Retrieval Augmented Generation (RAG) Using the Vertex AI Gemini API
• Lab: Introduction to Function Calling with Gemini

• Fundamental knowledge of machine learning, in particular generative AI, and basic understanding of security on Google Cloud.

None

None