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.
UnitedHealth Group's AI claim denial controversy exposed a systemic failure that most coverage missed: the underlying data the algorithm trained on was never independently scored for trust, provenance, or recency. When a payer deploys AI on unverified data to deny care at scale, the problem is not just the model. It is the data supply chain that no one audited.
HIPAA Safe Harbor removes 18 identifiers from health records, but modern AI can re-identify up to 99.98% of individuals using as few as 15 data attributes. The gap between what Safe Harbor protects and what AI can infer is the most dangerous blind spot in health data governance today.
The Change Healthcare breach exposed 100 million patient records and paralyzed claims processing for months. But the deeper failure was not the ransomware itself. It was the absence of any trust verification layer across health data infrastructure, meaning no one could confirm what data had been altered, accessed, or corrupted after the attack.
HIPAA defines two methods for de-identification, but neither method equals anonymization. The distinction determines whether health data can be re-linked to a patient, whether it qualifies for regulatory safe harbor, and whether AI models trained on it carry legal exposure. Most healthcare organizations conflate the two terms and absorb risk they never priced.
Mesothelioma kills over 2,500 Americans annually, with a median latency period of 30 to 50 years between asbestos exposure and diagnosis. Over 80% of cases are diagnosed at stage III or IV because the occupational exposure signals that predict risk sit in fragmented, unscored datasets that no clinical system connects. Behavioral intelligence offers a way to close that gap before patients present with advanced disease.
Most health system CTOs define AI trustworthiness by model accuracy, HIPAA compliance, and vendor certifications. None of those address the actual failure point: unscored, unverified training data flowing into production AI without provenance, consent validation, or recency checks. The trust problem is not in the model. It is in the data the model never should have seen.
Over 4,000 fabricated references were recently found across 2,800+ published research papers. When AI systems in healthcare produce outputs that cannot be traced back to verified source data, the consequences extend far beyond bad citations. Audit trails are the structural mechanism that makes health AI accountable, and most deployments still lack them.
Bad health data costs the U.S. healthcare system an estimated $3.5 trillion annually through redundant testing, failed care coordination, billing errors, and AI models trained on unverified records. The problem is not a lack of data. It is a lack of trust in the data that already exists.
Data poisoning attacks against health AI systems exploit the one layer most organizations never audit: the training data itself. A single corrupted dataset can shift diagnostic thresholds, suppress treatment recommendations for specific populations, or introduce systematic misclassification that persists through every downstream model. Without trust scoring at the point of data ingestion, health AI is building clinical decisions on an unverified foundation.
Over 350,000 health apps exist in major app stores, yet fewer than 2% have any clinical validation. Consumers trust their health app data far more than clinicians do, and this gap creates a structural problem for every organization trying to use patient-generated data in care decisions or AI training.
FHIR APIs promise patients the right to access and move their health data. The reality is that most patients can only retrieve a narrow slice of their record, and what they get lacks the provenance and trust scoring needed for meaningful use. The gap between legal right and practical access defines the next frontier of health data governance.
Employer wellness programs collect biometric screenings, mental health app usage, and claims data on millions of workers, yet most operate under consent frameworks designed for marketing opt-ins, not health data governance. The gap between what employees technically agree to and what they actually understand creates legal exposure, erodes trust, and poisons downstream data quality for any AI system trained on it.
Synthetic health data inherits, amplifies, and launders the biases present in real-world clinical datasets. When healthcare AI models train on fabricated records that were never scored for provenance or consent, the resulting bias becomes invisible to standard audits. The only defense is scoring data trust before any model trains on it.
Recency carries 15% of the Data Trust Index score, yet most health AI pipelines treat timestamps as metadata rather than a trust signal. Stale data silently degrades model accuracy, clinical decision support, and regulatory defensibility. Health data recency scoring is the single fastest way to separate actionable intelligence from archived noise.
The 21st Century Cures Act gave patients a legal right to access their electronic health data through standardized APIs. But access is not ownership. Without trust scoring, provenance tracking, and consent governance, the data patients receive is often incomplete, outdated, or unusable for the AI systems that need it most.
Head and neck cancer patients generate measurable behavioral signals weeks before their first oncology consult. These signals, from symptom searches to appointment delay patterns, represent a layer of patient intelligence that clinical systems never capture. Understanding this pre-clinical window changes how pharma, health systems, and researchers identify and reach patients earlier.
A patient's ZIP code predicts life expectancy more reliably than their genetic code. Geographic health data carries enormous weight in risk models, population health, and SDOH analytics, but most of it lacks provenance, recency, or consent verification. DataSpine scores geographic SDOH data across all 8 trust dimensions before it enters any model.
Most health data onboarding processes copy records into a new environment, creating compliance risk and provenance gaps before analysis even begins. IntegrityNet eliminates data movement entirely, scoring and cataloging EHR data in place through a zero-copy architecture that preserves chain of custody from the first connection. Here is what that process actually looks like, step by step.
The CMS ACCESS model ties $420 per beneficiary per month to outcome-aligned chronic care management, but participating organizations cannot meet reporting requirements if their underlying data fails on provenance, recency, or concordance. Health systems applying for ACCESS need a data foundation that goes beyond EHR completeness. This post breaks down what ACCESS actually requires, what most applicants underestimate, and how trust-scored data closes the gap.
Most health data platforms store records. Few make those records queryable by trust score, consent status, and provenance chain simultaneously. The Data Reservoir is the layer that converts DTI-scored health data into queryable intelligence, letting researchers, health systems, and pharma teams run queries that return not just answers but auditable confidence levels.
The DTI score assigns every health data record a grade from Bronze to Platinum across 8 weighted dimensions. Platinum-grade data is not a marketing label. It is a contractual guarantee that a record meets verifiable thresholds for provenance, consent, recency, quality, concordance, validation, breadth, and stability before any AI model or regulatory submission touches it.
Health systems that build data trust infrastructure in 2026 will own the competitive position that late movers cannot replicate. The window is narrow: regulatory pressure, AI deployment timelines, and payer contract shifts are converging on a single year. Institutions that wait until 2027 will spend more to catch up than early movers spent to lead.
Most healthcare organizations treat consent as a single checkbox. A five-tier consent architecture separates identity verification, data use authorization, purpose-specific permissions, downstream sharing controls, and revocation rights into distinct, auditable layers. ConsentOS operationalizes all five tiers so every health data record carries machine-readable consent metadata before any AI model touches it.
Most companies build a philanthropic arm after the money is in the door. We built Sean's Friends on Day One, before the company was profitable, because the order matters. Giving is the tenet. Profit follows the tenet.
