Most data governance was built to answer one question: can this person be identified?
That question is no longer enough.
AI systems don't just read information. They connect it. A support ticket, an email thread, a calendar history, a set of notes from three different teams — individually, each one is mundane. Linked and summarized, they become something else entirely.
You're no longer holding records. You're holding a version of a person.
It can be searched, compared, and reused in ways nobody specified when any single piece was first collected. And the person who agreed to share their information for one purpose did not agree to be assembled.
Consent to share is not the same as consent to assemble.
None of this is an argument against connecting data. Fragmentation causes real harm too, people repeating themselves endlessly, teams deciding without the full picture, work duplicated across silos.
It's an argument for knowing what you've built, and being able to explain it to the person it describes.
What this means for health organizations
Health organizations often hold information collected by different programs for different purposes. Before connecting it for AI search or summaries, map what's being linked, check it against what people agreed to, and limit who can query the combined view.
