Input Sanitization for Agentic Systems: What Actually Works

100 Queens Quay East, Toronto

Thursday, 6:30 PM → 8:00 PM

Agentic systems amplify every classic LLM safety problem. A prompt injection isn't a jailbreak, it's remote code execution by way of your assistant. A PII leak isn't a compliance footnote, it's training data for a vendor's next model. KC walks through the design of a production sanitization layer that sits between untrusted input and the model. Why generic guardrails fail. A layered detection model. Context-preserving pseudonymization that doesn't break downstream reasoning. Integration trade-offs with real latency numbers.

Speaker: KC Udonsi, DC416 Co-Organizer

RSVP Now