I like to build AI agents, break them, and teach others to do both.

Defences for AI that acts on its own

Models, platforms and research for keeping language models and agents safe in production. Most of it follows the same loop: build the system properly, attack it the way a real adversary would, then turn whatever breaks into controls that ship. That covers everything from a spoken prompt to a tool call, and the regulations an agent has to satisfy before a bank will let it run.

Leading AI security where the stakes are highest

I've led AI security for governments, architected platforms for regulated finance, and advised the founders and CISOs deciding how their organisations adopt AI.

I teach engineers to build AI, then to break it

I've taught at four universities, including two IITs, and run workshops across the UAE and India. My classes are hands-on: students build an agent in the first week and attack it by the last.

On stage across the Gulf and Asia

Keynotes, panels, moderation and workshops at security, energy and technology events, mostly on agentic AI threats, sovereign AI and responsible adoption.

Every stage, year by year

Fifteen keynotes, panels, workshops and lectures since 2020, across the UAE, India and Singapore. Most of them are about the same question: what changes when AI starts to act.

Books, standards and research

Two books written six years apart: the first on how to build perception models, the second on how to deploy agents safely.

From building models to securing them

Education, honours and certifications

A PhD in progress, degrees and executive study in AI and governance, and a long record of competitive ML.