Work · Regulated healthcare AI
Clinical-AI governance under active regulation
HTI-1 & state AI-law frameworks ยท 25% adoption lift
The context
Clinical decision-support AI sits in one of the most unforgiving settings for the technology: a wrong output can affect patient care, and the system answers to a regulator, a clinical-safety review, and a board at the same time. A platform in this space needed governance that would let it ship into real clinical use without becoming the reason it got stopped.
The mandate
Build the governance frameworks for a clinical decision-support platform navigating HTI-1 and the emerging patchwork of US state AI regulation, governance robust enough to satisfy a regulator, legible enough for a board, and practical enough for the people building and operating the system every day.
The constraints
The work had to hold under real enforcement, not aspiration: overlapping federal and state requirements, clinical-liability exposure, and sensitive patient data. And it had to do all of that without smothering the product, governance that blocks adoption is its own kind of failure. The bar was AI that is provably safe and usable.
The intervention and method
The frameworks anchored to recognized standards a regulator and a board both accept, then mapped the specific HTI-1 and state-law requirements onto the system's actual behavior, accountability for each AI-influenced decision, controls over data and model behavior, logging, and human checkpoints where the stakes demanded them. Every consequential behavior was stress-tested before deployment rather than after (Crucible), because in a clinical setting the cost of a wrong output is realized, not hypothetical. This is the operating model set out in AI governance for regulated institutions, applied under live regulation.
The outcome
The system was positioned to withstand regulatory and board scrutiny at once, and adoption rose 25%. That is the point the whole practice turns on: governance done right does not slow adoption, it unlocks it, because people trust and use what has been shown to hold up.
This is the case that most directly proves the thesis: AI that clears a regulator and a board is adopted faster, not slower.
What this demonstrates
AI governance & compliance · AI systems: build & run