Assuring an enterprise AI assistant: a prompt injection hidden in a retrieved invoice is treated as data, not instructions, and the assistant's attempt to email the customer list externally is blocked pending human approval; six defended layers — application, tools and agents, inference, model, supply chain and data — each with its own control; and a red-team result of 46 of 48 abuse cases mitigated, with two residual risks owned and reviewed, because guardrails are necessary but never sufficient.

CY-EXPERT-074Expert Level · AI Security

AI Security Assurance — Securing and Assuring AI Systems

A twenty-two-hour expert course where you assure an enterprise's AI systems across every layer — data, model, supply chain, inference, tools and agents, and the surrounding application — treating each guardrail as necessary but never sufficient, and recording residual risk honestly.

22 hours12 chapters48 lessons241 activities12 labs

What you'll learn

Map an AI system's attack surface and threat-model it into testable abuse cases.
Secure the data, model, supply-chain, inference, tool, and application layers with defense in depth.
Defend against adversarial evasion, prompt injection, and unsafe agentic action.
Red-team, continuously assure, and govern AI within risk appetite, honest about residual risk.

Price:৳12,000