Research

The through-line of my career is straightforward: I've spent over twenty years asking the same question in different domains. How do you build a system so that the trust you place in it is actually warranted?

The domains have shifted — microkernel operating systems, TPM-based trusted computing stacks, smartphone security, e-health infrastructure, software-defined networking, IoT gateways, and now autonomous AI agents. The methods have evolved. The underlying question hasn't. Each domain surfaced its own version of the same problem: adversarial input, unreliable output, unbounded execution, and the architectural discipline needed to contain all three.

That's why I keep saying AI security challenges aren't new — only amplified. Prompt injection is adversarial input. Hallucination is unreliable output. Agent misbehavior is unbounded execution. The vocabulary is new. The problems are the ones system security has grappled with for decades.

My published work sits at the intersection of two worlds most people work in separately: academic system security research and enterprise-scale security architecture. That intersection is where the Frameworks come from. PALIM formalizes design patterns I first saw as recurring failure modes across enterprise GenAI deployments — but the reasoning behind each component draws on years of research into adversarial ML, isolation architectures, and trusted execution. LAMINA does the same for the agentic control plane.

Every Framework I publish, every deep dive in the Writing section, and every talk in Media is grounded in the same discipline: translate research into architectural decisions engineers and CTOs can act on.

Publications — 50+ scientific and technical papers spanning operating system security, trusted computing, smartphone and mobile security, e-health infrastructure, SDN security, IoT security, and secure AI systems.

Books — Contributed chapters and edited volumes on system security topics.

Topics — Deeper reference material on the domain areas I've worked in: operating system security, trusted computing, smartphone security, e-health security, SDN security, and IoT security.

My active work focuses on security architecture for autonomous AI systems at enterprise scale — specifically:

• Deterministic policy enforcement for agentic AI, where the model itself cannot be trusted to honor rules.
• Identity, delegation, and access control models that survive when one human authorizes many agents acting on their behalf.
• The architectural boundary between the AI application and the enterprise environment it operates in — data isolation, tool containment, runtime observability.
• Threat modeling for GenAI and agentic systems, translating research (adversarial ML, LLM vulnerabilities, agent safety) into design patterns architects can apply.

These are the questions PALIM and LAMINA respond to. They're also the source of everything I write and speak about publicly. Research isn't something I did before this. It's what I'm still doing.