Foundations of AI-Based Application Exploitability
AI now sits inside the application. When a model reads a customerā€™s message, retrieves a document, or calls an API, it becomes part of the attack surface, and it canā€™t reliably tell instructions from data.
Every AI-enabled application has weaknesses. What matters is which ones an adversary can actually exploit, and whether your organization could prove it has them under control. This course gives you a shared language and a practical method for assessing, testing, and governing AI exploitability, using Aria, a fictional bankā€™s AI agent, as the case study throughout.
You'll learn:
- Separate vulnerability, exploitability, and risk, and know where CVSS, EPSS, KEV, and SSVC fall short for AI
- Map the attack surface of a modern application across APIs, cloud, the supply chain, and the AI layer
- Explain why AI systems are steerable: one channel for instructions and data, probabilistic behavior, and agency
- Trace AI attack chains from reach to impact, and find the deterministic controls that break them
- Rate AI findings, test the whole chain, and measure end-to-end success rates
- Close the AI ownership gap with three lines of defense, risk appetite, and board reporting
- Apply five control principles and a minimum control set for your role
- Know what evidence regulators and examiners expect, and what you could prove on day one
