Utah experts back synthetic data, stronger governance to balance privacy and public use

A panel of Utah lawmakers, university researchers and health-data officials met in Utah to discuss how synthetic data, stronger identity controls and new governance models could preserve individual privacy while allowing agencies and researchers to use data for public benefit.

The discussion focused on three linked questions: how long government should retain records, which transformation techniques (from redaction to differential privacy and synthetic data) protect people best, and what legislative or governance structures the state should build to make useful, privacy-preserving data available. Representative Lisa Shepherd said lawmakers must “listen to the constituents” while updating record-access rules to accommodate new privacy-preserving tools.

Panelists stressed that synthetic data — artificially generated datasets that preserve statistical patterns of original records but are not the original records — can let analysts test models and run research without exposing actual personal data. “Synthetic data as we define it refers to artificially generated data that preserves statistical characteristics and structural patterns in the data that reflect the original but are not the original,” said Dr. Rudolph, a panelist who described a COVID imaging project that used synthetic images so researchers could complete doctoral work without using patient-identifying records.

Samuel Smith, a member of the Utah State Privacy Commission, framed the technical challenge more broadly: “The fundamental problem with data is that it wants to be correlated,” he said, arguing that as ambient datasets and AI make reidentification easier, privacy protections must include economic and governance tools, not only technical fixes.

Panelists compared common transformation techniques. Zach Boyd, who runs the state’s office of AI policy, described de-identification, aggregation and differential privacy as approaches that trade some data fidelity for stronger reidentification resistance. He said newer AI-driven generative methods can create synthetic datasets that are useful in many analyses but that such outputs lack the formal theoretical guarantees that techniques like differential privacy offer.

Speakers gave concrete examples of synthetic data use. A university research team turned clinical X‑ray image sets into synthetic image data to permit COVID imaging research without exposing patient records. Another project described using protected synthetic health records to surface population-level signals that helped hospital systems identify risk patterns for self-harm without revealing individual patients.

Panelists also discussed how synthetic data changes operational risk and access models. One panelist said agencies can keep original records in tightly controlled “cold storage,” use them only in secured environments to train or retrain synthetic generators, then distribute the synthetic outputs to a wider set of users. That approach, the panel said, reduces the attack surface because the high-value original records are rarely accessed directly.

Identity, provenance and linkage of records were recurring themes. Several speakers proposed user-controlled linkage or “digital twin” models that allow individuals to authorize temporary, mission-limited joins across datasets for a specific service (for example, letting a physician see linked data during a clinical visit) while keeping broad population-level analytics based on synthetic data. “You can link them by providing a cryptographic provable assertion that says I belong to both datasets,” Samuel Smith said, describing an on-demand linkage model.

Panelists noted legal and policy constraints in state law. Several speakers referenced Utah’s Government Records Access and Management Act (GRAMA) and the state’s rules that classify communicable-disease information as especially sensitive; those constraints, they said, can limit researchers or public-health officials unless statutory updates or clarified guidance are adopted. Representative Lisa Shepherd said lawmakers should work with stakeholders and use interim committee time to learn how privacy-preserving techniques could be incorporated into GRAMA and related statutes.

On governance and financing, panelists proposed several models: a public-data “utility” held in trust, public‑interest oversight of synthetic‑data engines, and rules that restrict commercial actors (for example, insurers) from making individualized underwriting decisions using non‑aggregated government-held data. Several speakers argued that where governments allow licensed commercial reuse, compensation or microtransaction models might be appropriate so citizens and the public sector share value created from government datasets.

Panelists also flagged technology on the horizon. One speaker urged planners to track quantum computing as a future capability that could change reidentification and search risks, and several urged building auditable software and data supply chains so automated tools can enforce retention and use policies.

The panel closed with lawmakers and privacy commissioners saying Utah has momentum on privacy policy and should use stakeholder engagement and interim sessions to craft legislative changes. Representative Lisa Shepherd said the process will require more listening, interim work and stakeholder engagement before any bill is introduced.

The meeting recorded no formal votes or ordinance actions; panelists and legislators said follow-up discussions and interim committee work will determine next steps and potential statutory changes.