The foundational assumption underlying U.S. health data governance – that data are either "identified" and protected or "de-identified" and unregulated – is both legally incoherent and technically obsolete. Enacted when de-identification seemed a plausible safeguard, the HIPAA Privacy Rule codified this binary into law. HIPAA offers two de-identification pathways: expert determination and the safe harbor method. Institutional incentives have driven near-universal adoption of the safe harbor, producing , a form of legal endogeneity in which symbolic compliance with the eighteen HIPAA enumerated identifiers became mistaken for substantive privacy protection. The Common Rule's divergent identifiability definition has compounded this incoherence, and recent NIH attempts to contractually harmonize these conflicting standards represent regulatory surrender rather than reform.Re-identifiability is not a binary condition but a multi-dimensional, temporally shifting continuum shaped by the availability of external key data, evolving adversarial capabilities (including machine learning and AI), and the motivation of those seeking privacy breaches. Re-identification risk is also distributed inequitably, falling disproportionately on minoritized populations and individuals with rare conditions – precisely those whose data hold the greatest potential for advancing health equity. The binary regulatory framework not only fails to address this heterogeneity but actively reproduces its structural inequities.Drawing on Lessig's theory of regulatory modalities, we argue that privacy governance is an architectural problem as much as a legal one. Effective governance must redesign the architecture of data systems, not merely update their legal labels. We review novel technical approaches including k-anonymity frameworks, tiered access systems, controlled-access enclaves, and context-responsive suppression. We then begin to elucidate what level of re-identification risk is ethically acceptable for what data. Critically, the achievement of "privacy" cannot be resolved by technical thresholds alone.We ground these arguments in our experience designing the National Dementia Workforce Study (NDWS) secure data enclave, contrasting our approach with the All of Us Research Program. Both cases demonstrate that meaningful expert determination already occurs through iterative, multidisciplinary deliberation among lawyers, bioethicists, data scientists, and community representatives, and not through fixed methodologies.Federal regulators need not undertake the Sisyphean task of overhauling HIPAA to begin meaningful reform. What is required is a deliberate shift away from safe harbor compliance as a default, toward ML-enabled expert determination-grounded governance frameworks as adaptive and equitable as the risks they govern.