← All SANS courses
Course overview · SANS SEC536 · Co-author with Mick Douglas
Adversarial AI
Penetration Testing AI Systems
Assess AI applications by testing how their models, data, APIs, and tools behave under attack.
What you’ll work on
Inputs and retrieval
Test how attacker-controlled prompts and retrieved content affect an application’s behavior.
Models and infrastructure
Investigate weaknesses in model assets, data pipelines, and the APIs surrounding them.
Alignment and interfaces
Explore safeguard failures and attacks delivered through text, images, or audio.
Agents and MCP
Examine how tool access and delegated actions turn an input flaw into a wider compromise.
Skills to take back to your work
- Test trust boundaries across interconnected AI components.
- Record reproducible evidence of AI-specific weaknesses.
- Link findings to protective controls and detection.