AI security expert. I audit how you run your LLMs - guardrails, security harnesses, and the gaps   that let agents go sideways in production.