Breaking the Black Box: An Introduction to SEC536 Adversarial AI
An introduction to SEC536 and the attack surface of AI-enabled applications. The webcast explores how assistants can be manipulated and why conventional application testing needs to account for models, data, and delegated actions.
Session information at SANS. Sign-in may be required to access webcast materials.
Session topics
- Prompt injection, jailbreaks, model poisoning, extraction, and agent abuse.
- How manipulated assistants can bypass safeguards, disclose information, or perform unintended actions.
- Adding adversarial AI assessment to an offensive security program.