February 25-27, 2026
Montreal, Canada
Rogue LLMs: Securing Prompts and Ensuring Persona Fidelity
This session demonstrates how to detect, analyze, and prevent prompt leaks and persona failures in large language models. Participants will learn proven techniques and tools for securing AI prompts, enforcing consistent bot behaviour, and integrating evaluation and threat modelling into real-world engineering workflows.
Ben Dechrai
Venn Labs
Ben has spent more than 20 years helping companies improve how they build software, more recently the practical side of agentic engineering: getting tools like Claude Code to deliver production-quality work rather than demos, and structuring repos, prompts and workflows so the output is effective, resilient and secure. That rests on decades of building and shipping product in application security and identity. He is a Microsoft MVP and has given 100+ talks on six continents, including DEF CON.
Sponsored by
