// AI_SECURITY_ARCHITECT — v2025

Josh
Hill

Building the security architecture for AI systems that actually run in production. Zero trust → agentic pipelines → enterprise scale.

Security architect.
AI systems builder.

I'm a security architect at IGT, working across identity, endpoint, cloud, and application security — with a deepening focus on the emerging discipline of AI security architecture.

Most people discussing AI security are studying it. I'm building it: production agentic pipelines, MCP server architectures, and zero-trust patterns for non-human AI identities. The field is being defined in real time, and I'm writing from the inside.

My background spans zero trust design, Entra ID/identity federation, PKI, SIEM/SOAR, and cloud security across Azure environments. I've led security evaluations, built automation pipelines, and navigated the post-merger complexity of aligning two enterprise security stacks.

This site is where I share what I'm learning — for practitioners who want less theory and more architecture.

10+
Years in security
3
AI pipelines in prod
Zero Trust Architecture
Agentic AI Security
Identity & Access (Entra ID)
LLM Threat Modeling
Cloud Security (Azure)
Endpoint Protection
MCP / Tool-Use Patterns
OWASP LLM Top 10
PKI & CipherTrust
SSDLC / AppSec
MITRE ATLAS
AI Governance (NIST AI RMF)

From the architecture
floor, not the sidelines.

Agentic AI
Securing the Agentic Pipeline: Lessons from Production
How we built a ThreatLocker-to-Claude approval pipeline at enterprise scale — and what we learned about trust boundaries, tool misuse, and autonomy controls.
8 min read
MCP Security
MCP Server Architecture: Attack Surface You Haven't Considered
Model Context Protocol opens powerful integration patterns for AI systems. It also opens a new class of privilege escalation and tool-chaining risks that most teams aren't thinking about yet.
6 min read
Zero Trust
Zero Trust Doesn't Stop at the Human: Extending ZTA to AI Agents
AI agents call APIs, read files, and send messages. If your zero trust policy doesn't account for non-human identities with dynamic permissions, you have gaps.
5 min read

Things I've
actually shipped.

AI Software Trust Pipeline
Claude APIAzure FunctionsVirusTotalTeams
Automated hash enrichment and AI-reasoned trust scoring for ThreatLocker approval queues. Reduced manual review time and introduced confidence-scored policy recommendations.
● Production
MCP Server for Enterprise AI
MCP ProtocolAzureChatGPT Enterprise
Architecting an HTTP-exposed MCP server to bridge enterprise AI subscriptions with internal security tooling — with auth, scoping, and audit logging as first-class concerns.
● In Progress
SecureLink Consolidation
SecureLinkLegacy IGTLegacy Everi
Cross-merger remote access platform consolidation — normalizing vendor access controls, eliminating duplicate instances, and enforcing consistent least-privilege posture.
● In Progress

Let's talk AI
security architecture.

I'm interested in conversations with practitioners building or securing AI systems, security leaders thinking through AI governance, and anyone working on agentic architectures at scale.

I'm also open to advisory conversations, speaking opportunities, and exploring what's next in my career as this field evolves.

Based in San Marcos, TX → NC