What happens when enterprise AI governance meets the classroom

We spent a semester with USC Marshall's AI Capstone cohort. Here's what we built, what we learned, and why this collaboration matters for the future of accountable AI.

Spring 2026  ·  nxtlinq Team  ·  AI Governance  ·  Education

A few months ago, we kicked off something that didn't fit neatly into any of our usual playbooks. No product sprint. No pilot agreement. No go-to-market motion. Instead, we walked into a classroom at USC Marshall School of Business and started a semester-long conversation about the thing we obsess over most: how do you actually govern AI systems operating inside real business workflows?

The program — the USC Marshall AI for Business Capstone — was designed precisely for this kind of collaboration. Sponsored by Charlie Hannigan, it connected enterprise operators with students ready to wrestle with real problems, not sanitized case studies. We were lucky to be one of them.

Where we started

The first sessions were grounding exercises. Before anyone wrote a line of code or stood up an agent, we had to confront a question the industry has largely been dancing around: traditional Identity and Access Management (IAM) was built for humans. So what happens when the entity requesting access, making decisions, and taking actions is an autonomous AI agent?

That's the gap nxtlinq was built to address. Students were introduced to concepts like Human Identity Tokens (HIT), AI Identity Tokens (AIT), and how immutable, blockchain-anchored execution records create the kind of auditability that enterprises — and their regulators — will eventually demand. These weren't theoretical lectures. They were the scaffolding for everything that came next.

“Enterprises are quickly moving beyond simply deploying AI models. The next challenge is operationalizing AI responsibly.”

Getting hands-on

Once the conceptual foundation was in place, the semester shifted into implementation. Students stood up agents, connected them to data sources, reviewed execution logs, and traced the full automation arc: data → analysis → decision → action. It's one thing to describe that chain on a whiteboard. It's another to watch it play out in a live environment and realize how quickly accountability disappears when there's no identity layer underneath it.

Teams then took ownership — designing and deploying their own automation agents with defined governance boundaries and oversight models. We didn't shy away from the uncomfortable parts either. Governance stress testing included unauthorized access simulations and privilege escalation exercises. The goal wasn't to break things. It was to help students think beyond "does this agent work?" toward "how do we know it stayed within its authorized scope?"

The feedback we didn't expect

One of the more valuable — and honestly surprising — outcomes was the direction the feedback flowed. Yes, we shared how the platform works. But students evaluated our dashboard observability, proposed new analytics concepts, and framed nxtlinq's architecture from the perspective of enterprise stakeholders they'd never actually been: CIOs, CISOs, AI Governance Officers.

Key observation: The next generation of builders is approaching AI not just as a productivity tool, but as infrastructure — one that will require identity, policy, and accountability layers as autonomous systems become more deeply embedded in enterprise operations.

That framing sharpened things for us. When you see your own platform through the eyes of someone who just spent a semester trying to break it, audit it, and justify it to an imaginary CISO, you come away with a cleaner sense of what actually matters.

What we're taking forward

This collaboration reinforced something we've been watching across the enterprise market: the conversation has shifted. It's no longer about whether to deploy AI. It's about whether you can answer for what your agents do — who authorized them, what they accessed, how decisions were made, and where the audit trail lives.

The students we worked with this semester already think this way. That's encouraging. And it's a signal that the industry's expectations around AI accountability are going to move faster than a lot of organizations are currently planning for.

We're grateful to the students and faculty at USC Marshall for making space for this kind of collaboration, and to Charlie Hannigan for building a program where these conversations can happen in a real, hands-on context. We're looking forward to what comes next.

nxtlinq — Governed AI, from identity to execution