Duration: 3 Days
Format: 100% Theory & Strategic Design
Level: Foundational

This course focuses on the core attack surfaces, supply chain risks, output security, and governance frameworks involved in modern AI deployments. Participants will learn how AI systems can be attacked, how those threats can be mitigated, and how structured risk management can be applied to enterprise AI initiatives.

Course Outline

Day 1: The AI Attack Surface & Prompt Vulnerabilities

Objective: Deconstruct the technical components of an AI system and identify where they are most vulnerable to manipulation.

Morning: Mapping the AI Attack Surface

• Training data risks, including data poisoning and privacy leaks during pre-training.

• Model weights security, including risks of extraction or unauthorized modification.

• Inference pipeline vulnerabilities such as API security issues and resource exhaustion.

• Agent tool use and the high-risk permission gap when agents interact with files, databases, and APIs.

Afternoon: Hijacking Intent (Prompt Injection)

• Direct prompt injection and attempts to override system instructions.

• Indirect prompt injection through malicious content hidden in websites, emails, or PDFs.

• Adversarial suffixes and how non-human-readable tokens can bypass model safety filters.

Day 2: Supply Chain Risks & Output Security

Objective: Analyze the dangers of the AI ecosystem and implement defensive guardrails.

Morning: Poisoning & Supply Chain Integrity

• RAG store vulnerabilities and the risk of poisoned documents in vector databases.

• Third-party model weights and the dangers of downloading backdoored models.

• Dependency risks in AI development libraries and related packages.

Afternoon: Output Sanitization & Guardrails

• Guardrail architectures using checker models to validate outputs.

• Data Loss Prevention (DLP) strategies for identifying and redacting sensitive information.

• Jailbreak detection using semantic analysis to identify harmful or unsafe outputs.

Day 3: AI Governance & Risk Frameworks

Objective: Operationalize AI security using industry-standard frameworks.

Morning: The NIST AI Risk Management Framework (RMF)

• Govern: Establishing organizational policies and a risk-aware culture.

• Map: Identifying the context and risks of each AI use case.

• Measure: Testing, evaluating, and benchmarking AI systems.

• Manage: Prioritizing and addressing risks to reduce impact.

Afternoon: Strategic Implementation & Compliance

• Incident response planning for jailbroken models or data leakage events.

• Threat modeling for AI agents with tool access.

• Emerging AI regulations and their impact on corporate risk strategies.

Key Learning Outcomes

By the end of this course, participants will be able to:

• Identify and categorize vectors of attack across the AI lifecycle.

• Design a defense-in-depth architecture with input and output guardrails.

• Apply the NIST AI RMF to create a structured governance plan for enterprise AI deployments.

• Recognize key AI supply chain risks and output security concerns.

• Understand how to prepare incident response and compliance strategies for AI systems.