About
Christopher M. Graziul, PhD
Computational sociologist. Independent advisor on AI governance durability.
What I do
I'm a computational sociologist who works on a single problem: how long does AI governance last before something breaks. Most frameworks measure compliance at a moment. I measure durability — how governance survives model cycles, vendor shifts, regulatory change, and staff turnover.
My background is structural analysis of high-stakes data systems. PhD in sociology from the University of Chicago; applied math and physics undergrad at Virginia Tech; postdoctoral work at Brown. Before launching this practice I was a research assistant professor and data scientist at the University of Chicago. My position paper at ICML 2025 on AI governance came out of that work, not ahead of it: the theory is what running the data forced me to figure out.
I learned durability the hard way, running an NIH-funded research program on sensitive data. I built a first-of-its-kind speech corpus, wrote five annotation protocols for more than thirty annotators, and drafted the data-use agreement myself because no template fit. The records were already public and still deserved real protection, so I held them to a standard the law did not require: sensitive by choice, not by statute. That work became the Sensitive Open Data framework.
I write governance frameworks and build decision-process architectures that survive institutional change. The gap between what NIST AI RMF and ISO 42001 ask organizations to do, and what actually happens when models update or the team that wrote the policy leaves: that gap is where the work lives. I call it governance shelf-life.
I also founded the Illinois Data Equity Project, a 501(c)(3) whose founding governance is designed to fold itself into community leadership over time. It is the methodology applied to its own institution. Read the case study →
Selected work
-
ICML 2025 position paper on AI governance
Theory pulled from practice: the structural durability of AI governance frameworks across model and regulatory transitions, generalized from years of running high-stakes data systems.
-
Peer-reviewed research
Computational social science, urban sociology, social-network analysis, and applied NLP. Google Scholar.
-
Research Assistant Professor & Data Scientist, University of Chicago (2018–2025)
Computational social science, including translational AI work and methodological infrastructure for structured data systems.
Full academic record at graziul.github.io.
How I work
What you can expect from working together.
Specific over abstract. Roles are paired with named people. Triggers are quantified ("when vendor pricing changes by more than 20 percent"), not vague. Everything has a date or an owner.
Concession over salesmanship. I will tell you when this methodology does not apply, when a competitor is better positioned for your specific situation, or when you don't need an engagement at all.
Productized where it makes sense. Tier 2 has fixed prices, fixed scope, fixed timelines. Hours aren't a useful unit when the work is structured analysis.
Slow on purpose where slow earns its place. The free durability check is instant and self-serve. The paid Shelf-Life Assessment is the one that takes seven to ten business days, because a real human reads it. Same-day depth is impressive but it is not what the paid work is.
Curious whether this applies to your situation? Start with the diagnostic.