Essays & technical papers
I write about the design decisions that make AI systems trustworthy by construction: architecture, compliance, and the ethics of automating decisions about people.
Constraints, safety, and provenance are load-bearing structure, not paperwork bolted on after launch. A design stance for building AI systems that cannot be configured into failure.
A lawyer who evaluates evidence for a living believed a voice that was never there. The federal government now counts cases like his. It counts only the ones the victim figured out.
Proving you are a real person used to be free and assumed. It became a task, then a product, and the people carrying the most of it have the least room to refuse.
The fear was that a fake would be believed. The sharper risk is that something real stops counting as proof. American courts have now heard that argument several times. It has lost every time, and the reason it lost is not reassuring.
The data on AI vendor risk tells a less dramatic, more useful story than the sales pitch: a real coverage gap, real procurement friction, and very little evidence anyone is writing large new checks.
Part I of III. Orbital AI data centers don't have a bandwidth problem, they have a heat problem: a thermodynamic law no one can bargain with, hiding one layer beneath the laser downlinks everyone argues about.
Part II of III. Going to orbit doesn't subtract a regulator, it stacks several incompatible sovereigns onto one moving object, and the seams between their legal theories are where accountability dies.
Part III of III. The governance gap orbit creates is filled by nothing but operator discretion, and the analysts we'd use to check that discretion are paid by the same people whose discretion is in question.
A generational fault line is forming around synthetic media: one cohort adapts to a post-authentic internet while another retreats into smaller, verified circles. What that split means for platform trust, and how long the window for intervention stays open.
AI governance keeps landing on program managers' desks whether the org chart says so or not. A case for treating the discipline as architecture, not administration.
Inside an open-source terms-of-service risk engine built to run entirely on local, on-device models, and what it means to make a governance idea operational.
A reflection on leading AI governance for a hiring system used across a large multi-brand retail footprint, and what consent, bias, and dignity look like from the applicant's side of the screen.
