Results from nearly 30% of clinical trials deemed highly likely to require mandatory reporting have never been submitted to the federal government’s public database, according to an internal U.S. Food and Drug Administration (FDA) analysis disclosed alongside a regulatory enforcement push launched in April 2026.

The FDA sent reminder letters to more than 2,200 companies and researchers, warning that failure to report clinical trial results to ClinicalTrials.gov may result in financial penalties. The letters correspond to more than 3,000 registered trials, a subset of which received public funding. That 30% non-submission figure cuts at a problem medical researchers have flagged for years: when trial results don’t reach the public record, independent replication collapses, and clinicians can’t assess whether a drug’s mechanism holds under scrutiny.

This is a reporting crisis. Not a paperwork problem.

The gap between what gets registered and what gets reported has drawn sustained criticism from academic researchers who argue that selective publication distorts the evidentiary base on which prescribing decisions rest. A trial registered but never reported may have produced null results, adverse signals, or data that contradicts a sponsor’s preferred narrative. Absent mandatory submission to a federal repository, those findings disappear. The FDA’s own internal review confirmed what independent watchdog groups have documented repeatedly: the problem isn’t marginal, and it isn’t getting better on its own.

Under the Food and Drug Administration Amendments Act of 2007, sponsors of applicable clinical trials are required to submit results to ClinicalTrials.gov within 12 months of the primary completion date. The law carries civil monetary penalties for non-compliance. What’s been missing, critics said, is consistent enforcement. The April 2026 letters represent at minimum a signal that the agency intends to treat the statutory deadline as binding rather than advisory.

The agency’s acknowledgment of a long-standing researcher complaint deserves attention. For years, investigators attempting to replicate published findings have hit walls because raw data, protocol amendments, and negative outcomes don’t make it into the public record. Without that granular disclosure, it’s not possible to determine whether a reported hazard ratio reflects the full enrolled population or a curated subset. That failure to replicate doesn’t just slow science. It means physicians prescribing based on published efficacy data may be working from an incomplete picture, particularly when subgroup outcomes in populations such as Pacific Islanders, Native Hawaiians, or other underrepresented groups are buried or simply never submitted.

Replication is not optional. It is the mechanism by which clinical evidence earns the weight it’s given in guideline recommendations and treatment decisions.

The FDA’s enforcement posture arrives at a moment when the agency is also grappling with pressure to accelerate drug approvals and reduce regulatory burden. Those competing priorities produce tension worth naming directly: transparency requirements and speed-to-market incentives don’t always pull in the same direction. Sponsors who face fines for late submission still face no immediate consequence for reporting only favorable outcomes in summarized form, provided they meet the technical filing deadline. The 2026 reminder letters address the “submit anything” problem. They don’t resolve the “submit everything meaningful” problem.

This distinction matters for readers of clinical trial literature. Submission of a results record to ClinicalTrials.gov satisfies the statutory requirement. It doesn’t guarantee that the submitted record contains the endpoint granularity necessary for independent analysis. Overall survival (OS) data, subgroup hazard ratios, confidence intervals, and pre-specified versus post-hoc analyses are not uniformly disclosed even in submitted records. The FDA’s push, welcome as it is, addresses the first-order failure. The second-order failure, selective or incomplete reporting within submitted records, remains an open regulatory question.

Meanwhile, Novo Nordisk announced a partnership with OpenAI in April 2026, agreeing to integrate the ChatGPT maker’s artificial intelligence models across the Danish drugmaker’s operations. The stated goals include helping Novo Nordisk’s scientific workforce analyze complex datasets and compressing the timeline between early-stage research and patient-accessible treatment. Pilot programs are expected to launch initially in research and development, manufacturing, and commercial operations, with full integration targeted before the end of 2026.

The deal is the latest in a wave of pharmaceutical-AI collaborations that have accelerated since 2024, as major health care companies move to embed machine learning into drug discovery pipelines. Novo Nordisk’s decision to partner with OpenAI rather than build proprietary models internally reflects a strategic bet that commercially developed large language models can be adapted to pharmaceutical research workflows faster than in-house development would allow.

The clinical implications of AI-assisted drug discovery aren’t yet measurable in the endpoints this journal prioritizes. No OS curves exist for molecules that haven’t reached trial. No hazard ratios accompany a press release describing a partnership structure. Readers should calibrate enthusiasm accordingly.

What can be said is structural. AI platforms applied to dataset analysis may help identify biomarker correlations that human researchers working through conventional methods would miss or take considerably longer to find, particularly across large, heterogeneous patient populations where signal detection requires processing volumes of genomic, proteomic, and imaging data simultaneously. Whether those upstream efficiency gains translate into approved treatments, and into treatments that work in populations historically underrepresented in trial cohorts, depends on what data those models are trained on and what endpoints drug developers prioritize when they get to the trial stage.

That brings the two stories together.

If Novo Nordisk and OpenAI accelerate the timeline from target identification to clinical trial, but the resulting trials don’t report their results to ClinicalTrials.gov within the legally required window, or report them in forms too thin to permit replication, then the efficiency gain is captured entirely by the sponsor. The scientific community gets nothing it can verify. Regulators see a submission. Researchers see a summary. Physicians see a label. The underlying data, the kind that would let an independent team test whether an AI-identified biomarker actually predicts response in a Native Hawaiian patient population, doesn’t make it to the public record.

The FDA’s 2,200 reminder letters don’t fix that. But they establish, at least on paper, that the agency is watching the submission gap.

Reporting from STAT News on April 14, 2026, first brought together the FDA transparency enforcement action and the Novo Nordisk-OpenAI announcement, situating both within a broader moment of regulatory and commercial pressure on pharmaceutical development timelines.

The 30% non-submission rate is the number that should follow both stories. It represents trials where enrolled patients accepted risk, where public or private resources were spent, and where clinicians and researchers got nothing they can use. Whether the companies receiving reminder letters comply, and whether compliance produces records of sufficient quality to support independent replication, will say more about the FDA’s seriousness on this front than the letters themselves do.

Three thousand trials. Mandatory reporting requirements. The clock’s been running.