FDA Regulatory Capture: PDUFA Funding, AI Bypass, and the Speed-Safety Tradeoff

The FDA is accelerating approvals through two different mechanisms: better science and lower evidentiary requirements. The critical distinction is whether improved methods replace traditional evidence or regulators simply require less of it.

The Scientific Substrate Is Real

Validated biomarkers, surrogate endpoints, adaptive trials, master protocols, and improved real-world data can shorten development without proportionally reducing rigor.

Accelerated approval has also become more accountable. For oncology indications, average time from accelerated to full approval fell from 4.3 years in 1992–2013 to 2.3 years in 2014–2024. Average time to withdrawal fell from 9.5 to 3.2 years. The system has become faster at confirming successful drugs and removing failed ones.

The Discretionary Acceleration Layer

A separate acceleration push is political rather than scientific.

In February 2026, Commissioner Makary and then-CBER director Vinay Prasad proposed replacing the traditional two-trial standard with a single pivotal trial as the default for most drugs, without formal guidance or rulemaking.

The Commissioner’s National Priority Voucher program also proposed compressing selected reviews from 10–12 months to roughly one or two months under loosely defined national-priority criteria. This gives agency leadership substantial discretion over which sponsors receive the fastest treatment.

Both initiatives are unstable. Makary resigned on May 12, 2026 after a year with six drug-center directors. Prasad also departed. Without formal codification or their principal advocates, the single-trial standard and voucher program could be narrowed, delayed, or abandoned.

HHS OIG’s January 2025 review identified concerns in 3 of 24 accelerated-approval drugs examined. These included unplanned analyses and approvals granted despite reviewer or advisory-committee reservations. Two products were later removed from the market.

The result is a two-layer system: durable efficiency gains from better science, plus an unsettled attempt to accelerate reviews through administrative discretion.

The PDUFA Funding Question

User fees account for nearly 51% of FDA’s FY2026 enacted program level. Their share of PDUFA program costs rose from 7% in FY1993 to 77% in FY2025. PDUFA fees now fund more than 65% of the Human Drugs Program.

The FY2027 request would increase user fees by 7.7%. From FY2021 through FY2026, appropriated funding grew 7%, while user-fee revenue increased more than 39%.

This structure has two interpretations:

The EMA derives an even larger share of its operating budget from user fees, so the model is not uniquely American.

PDUFA, MDUFA, GDUFA, and BsUFA expire by the end of FY2027. Their renegotiation will determine FDA funding, authority, fee formulas, and exemptions for years. PDUFA VIII is therefore a live variable in any forward model of FDA behavior.

The AI Digital-Health Bypass

More than 96% of AI-enabled medical devices enter through the 510(k) pathway. Instead of proving safety and effectiveness through new clinical trials, manufacturers generally establish substantial equivalence to an existing predicate device.

A peer-reviewed study of 903 FDA-listed AI/ML devices found that almost one quarter reported no clinical performance study, while another 20% did not specify whether one occurred. Among devices reporting clinical evaluation, retrospective designs were most common.

Post-market failures track this validation gap. Roughly 43% of AI-device recalls occurred within one year of authorization, and most recalled devices had not undergone a clinical trial.

Disclosure is also weak. An analysis of more than 1,000 FDA summary documents produced an average transparency score of 3.3 out of 17. Evidence that cannot be independently audited provides limited protection.

Predetermined Change Control Plans allow manufacturers to define acceptable post-clearance algorithm updates without filing a new 510(k) for every change. FDA had authorized more than 1,350 AI-enabled devices by early 2026, roughly twice the 2022 total. PCCPs address model drift after clearance, not the initial validation deficit.

FDA is also applying generative AI internally. Elsa 4.0 supports a pilot using direct clinical-trial data feeds to accelerate analysis and clearance. Reports that Elsa produced fabricated or hallucinated studies expose a new failure surface inside the review process itself.

Digital therapeutics are the main exception. Products making explicit therapeutic claims generally require clinical evidence, often randomized controlled trials. The bypass is concentrated in diagnostic and clinical-decision-support software.

The Actual Speed-Safety Tension

The relevant divide is not fast versus slow. It is better evidence substitution versus evidence reduction.

That transfer is dangerous when surveillance capacity is already strained. FDA experienced six drug-center directors in one year, a reported 65% decline in advisory-committee meetings before partial recovery, workforce reductions, and at least one missed PDUFA goal date attributed to workload and limited resources.

Research on advisory committees does not strongly support the simple claim that financially conflicted reviewers routinely vote for industry. The stronger capture model is institutional: industry funds much of FDA’s operations, negotiates its legislative mandates, recruits former officials, and benefits from discretionary acceleration programs. Systemic bias does not require individual bribery.

The operational question is therefore not whether FDA is corrupt. It is how much current acceleration can be explained by stronger science, and how much requires funding or leadership incentives that push approvals beyond what the evidence independently supports.

Strategic Implications

The FDA contains three simultaneous forces: scientific progress, discretionary political acceleration, and industry-dependent funding. The analytical edge lies in separating companies benefiting from better evidence from those benefiting because less evidence is required.