SANS San Antonio 2026 - SANS@Night: Breaking the Black Box: An Introduction to Adversarial AI
A SANS@Night introduction to assessing AI applications, with demonstrations of assistants being redirected, exposing confidential data, and taking actions outside their intended scope. The session connects these failures to practical security assessments.
Session information at SANS. Sign-in may be required to access webcast materials.
Session topics
- How prompt injection and jailbreaks redirect an AI assistant’s behavior.
- Attack paths involving training data, model extraction, and agent capabilities.
- Where AI-specific testing belongs in a broader security assessment.