Answers are grounded in this paper and cite their sources.
AI products expose the same interface under contracts that give client data opposite confidentiality treatment. After a federal court denied privilege to consumer-AI exchanges because the platform's terms disclaimed confidentiality, this paper asks whether a clinician or a lawyer can even determine what a tier does. A reproducible governance-state audit represents each tier as a machine-checkable vector, measures disclosure reachability and reconstruction complexity, classifies documented boundary enforcement, and tracks declarations over time. The August 2026 six-tier, five-vendor audit found governance terms in 13/25 target-term pairs across five measurable crawl targets versus 20/20 commercial pairs, using unmatched vocabularies. The corpus spans 20,926 words and 59 conditions; all ten controlling documents showed normalized-content changes (median rounded window/change ratio: 25.5 days). The paper reads the first US AI-privilege rulings as a specification, names a proposed response (BAA-adjacent architecture), and derives requirements for visible, enforceable, versioned governance seams. Author preprint. This manuscript is under review and has not been accepted for publication.
No takes yet. Share an insight, caveat, or question.
Anonymous (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: