Insights
Data trust, provenance, and what AI should be allowed to act on. Written by Jason Alan Snyder, nearly every day. Longer essays live on his Substack.
Most explainable AI frameworks explain model decisions but ignore whether the training data was trustworthy in the first place. Glass box AI in healthcare requires more than interpretable algorithms. It requires explainable data provenance, where every record carries a verified chain of custody before a model ever touches it.
Most healthcare data platforms solve for infrastructure trust: uptime, encryption, access controls. Almost none solve for data trust: whether the records themselves are accurate, consented, current, and traceable. This distinction explains why billions in health AI investment still produce unreliable outputs.
Most health data architectures assume that sharing data means moving it. Zero-copy architecture breaks that assumption by allowing queries to execute where data already lives, eliminating duplication and the trust violations that come with it. But zero-copy alone does not solve the trust problem unless every record queried carries a verifiable trust score.
The ONC information blocking final rule forces health data to flow. But flowing data is not trustworthy data. Without a trust layer that scores provenance, consent, and quality before AI models consume health records, interoperability becomes a pipeline for unreliable inference.
Federated health data networks keep records behind institutional firewalls, but they do not solve the trust problem. Every node in a federation contributes data of unknown quality, unknown provenance, and unverified consent status. Without a trust layer underneath, federated learning in healthcare inherits every flaw of the data it never moves.
Explainability techniques tell you why a model made a decision, but they cannot tell you whether the data behind that decision was accurate, consented, or even real. The AI industry has spent billions making outputs interpretable while ignoring the fact that 91% of health data quality issues originate before a model ever trains. Trusting a model's reasoning is meaningless if nobody verified the inputs.
Over 150 million Americans wear a health-tracking device, but almost none of that data meets clinical-grade trust standards. The gap between consumer wearable output and clinically useful biomarker intelligence is not a technology problem. It is a data trust problem, and it maps directly to provenance, consent, and validation failures.
TEFCA will connect every major health network in the US through a common trust framework, but connectivity without data integrity creates new risks. Health systems that prepare only for technical compliance will find that TEFCA interoperability exposes every upstream data quality problem they have been ignoring.
Data quality measures whether a record is accurate and complete. Data trust measures whether that record should be used at all. Healthcare AI needs both, but the industry has invested almost exclusively in quality while ignoring trust, and that gap is where clinical AI fails.
Substance use disorder records carry federal protections under 42 CFR Part 2 that exceed standard HIPAA rules. Most AI systems training on health data have no mechanism to detect, segment, or honor these restrictions, which means SUD data either gets excluded entirely or mixed in without proper consent governance. Both outcomes damage model accuracy and patient trust.
The FDA's evolving guidance on AI in healthcare is converging on a single requirement most developers are not ready for: data provenance. Draft frameworks for AI-enabled medical devices and drug development now treat training data lineage as a core submission element, not an afterthought. Here is what is coming and what developers need to build now.
Over 90% of U.S. hospitals run on Epic or Cerner, yet no standard process exists to score the trustworthiness of EHR data before it enters an AI training pipeline. Without a trust score, models inherit every documentation gap, coding inconsistency, and consent violation baked into the source record. The Data Trust Index changes that by scoring every record 0-100 before any model touches it.
SuperTruth has published peer-reviewed research on the Data Trust Index (DTI) at Zenodo, making the formal methodology for scoring health data integrity publicly available. The paper defines the 8-dimension framework, weighted scoring model, and trust tier classification system that scores every health data record 0 to 100. This is the first open-access publication of a structured trust scoring methodology purpose-built for health AI governance.
SuperTruth now lets you score a health data record live on supertruth.ai. The DTI pipeline evaluates every record across 8 trust dimensions and returns a 0-100 score before any AI model touches it. This post explains what happens inside that pipeline, step by step, and why it changes how health data enters production.
Genomic data powers precision medicine, but fewer than 30% of clinical genomic datasets carry full provenance metadata. Without knowing where a variant call originated, how a sample was processed, and whether consent covers secondary AI use, precision medicine operates on assumptions rather than evidence. Genomic data trust requires provenance infrastructure that most health systems have not built.
Ninety-seven percent of providers agree that strong data strategies are essential for value-based care, yet most VBC programs still operate on claims data riddled with coding inconsistencies, missing fields, and stale records. Without a trust layer that scores every record before it enters a payment model, value-based care becomes volume-based guessing with extra steps.
Provenance carries the highest weight in the Data Trust Index for a reason: it is the dimension most likely to be broken, forged, or simply missing across health data systems. A single patient record can pass through 17 or more systems before reaching a decision point, and each handoff creates an opportunity for the chain of custody to fracture silently.
Fewer than 6% of the estimated 10,000 rare diseases have an FDA-approved therapy, and fragmented registry data is a primary reason. Rare disease registries collect critical patient information, but without structured data trust requirements, most of that information cannot reliably support research, regulatory submissions, or AI model training.
Most hospital systems rushing to deploy AI lack the data trust infrastructure that determines whether models succeed or fail in production. Readiness is not about compute power or vendor selection. It is about whether your data can be scored, governed, and trusted before a single algorithm touches a patient record.
HIPAA protects against unauthorized disclosure of health information, but it says nothing about whether that data is accurate, current, or trustworthy. The gap between HIPAA compliance and actual data trust is where health AI fails, clinical decisions go wrong, and patients lose confidence in the system.
Mental health data carries legal, clinical, and social risks that no other healthcare data category matches. Federal law treats it differently, patients fear it differently, and AI systems trained on it fail differently. Behavioral health data governance requires a consent architecture that most health AI companies have not built.
Roughly 80% of health outcomes are driven by factors outside the clinical setting, yet most health systems collect structured data on fewer than half of those factors. Health equity data measurement requires capturing what traditional EHRs were never designed to see: housing instability, food access, transportation barriers, and the trust dynamics that determine whether patients share information at all.
The average patient's health data is scattered across 19 different systems, and no integration standard has fixed the problem. Siloed health data costs U.S. healthcare roughly $150 billion per year in redundant testing, care delays, and administrative waste. The issue is not a lack of interoperability standards; it is that the data itself has never been scored for trustworthiness before anyone tries to connect it.
Children cannot consent for themselves, and their parents cannot consent forever. Pediatric health data governance sits in a legal and ethical gap that most health systems, AI developers, and researchers fail to address. The consequences show up in broken consent chains, unusable training data, and adolescents who avoid care entirely.
