Security Patterns

These are the patterns people reach for when they want to "secure" their AI system—input sanitization, output validation, guardrails, rate limiting, audit logging.

Some of these work. Some are theater. Most are necessary but not sufficient.

We document each pattern honestly: what it actually protects against, what it doesn't, and how to layer defenses effectively. No pattern here is a silver bullet. Used together correctly, they provide meaningful security.