Getting Started with Prompt Injection

AI systems are transforming how organizations operate, but they introduce an entirely new attack surface. In this session, SANS faulty fellow and author Joshua Wright will explore prompt injection, the #1 risk on the OWASP GenAI Top 10, and demonstrate how attackers manipulate Large Language Models into bypassing security controls and leaking sensitive data.

In this session you will learn about the controls used by language models to protect against misuse, and how attackers manipulate them using multiple techniques including keyword encoding, refusal suppression, format switching, and role play attacks. You will see how prompt injection exposes system prompts, RAG data, and agent integrations, and how it can enable classic injection attacks like command injection and SQL injection through AI interfaces.

At the end of the session we will examine defensive strategies including prompt hardening, multi-chain analysis, Human-in-the-Loop review, and prompt logging for threat hunting. Whether you are building AI systems or securing them, this session will give you the attacker's perspective you need to defend effectively.