AI Security & Trust

Protect your AI systems from prompt injection, jailbreaks, and data leaks.

Every production AI system is a potential attack surface. Prompt injection — where malicious input overrides your system prompt — is the SQL injection of AI systems, and it's already being used to attack real products. The OWASP LLM Top 10 documents the most critical vulnerabilities in LLM-based applications, and most developers aren't aware of most of them.

This track covers the full AI security landscape: prompt injection at the application and indirect levels (where malicious content in retrieved documents hijacks your agent), PII redaction with Microsoft Presidio (stripping sensitive data before it hits the LLM), red teaming your own models with Garak and PyRIT, and the zero-trust architecture patterns that production AI systems need.

AI security is not optional for production systems. If your AI agent has access to tools (database queries, email sending, code execution), an attacker who can manipulate its context can cause real damage. This track gives you the adversarial mindset and the practical defenses to build systems that are robust under attack.

11 Guides in This Track