At Pfizer, I authored the enterprise AI Security Plan and AI security controls, and I’m building the AI asset inventory and visibility layer. I attack LLM guardrails and agentic platforms before the enterprise relies on them.
I built a reproducible prompt-injection corpus with expected-safe outcomes so guardrails and model endpoints can be scored pass or fail. It’s now the standard harness for AI control evaluation.
I evaluated agentic SOC platforms and AI firewall and guardrail products against a defined AI security control set, informing which controls the enterprise adopts. I also devised a method to detect unsanctioned AI use from a public catalogue of AI application signatures.
Earlier, I built and productionised ML and deep-learning threat-detection models at Pfizer, including threat fusion with PU learning and pattern mining. My PhD research at the University of the Aegean focused on machine learning and multi-objective optimisation over big data; I have seven publications.
