Research indicates that high-value algorithms developed in academic settings often fail to diffuse to other hospitals due to the absence of a clear regulatory pathway for cross-institutional use. This bottleneck has created a landscape where groundbreaking tools remain siloed within the institutions that birthed them, limiting the broader societal benefits of medical innovation. Former FDA Commissioner Scott Gottlieb, M.D., recently addressed this stagnation by proposing a specialized federal regulatory framework in JAMA Health Forum. The central challenge involves the current oversight vacuum for “in-house” clinical AI, which is frequently managed as an internal clinical workflow rather than a regulated medical device. This lack of formal structure makes it difficult for health systems to share validated models, as the legal and safety risks of deployment outside the originating hospital are poorly defined. As academic centers increasingly rely on predictive algorithms, a standardized approach has become a critical priority for clinical integrity.
Addressing the Inconsistencies: The Current State of the AI Market
The Risks of Internal Oversight and Market Disparity
Currently, academic medical centers utilize vast repositories of electronic health record data to develop and deploy predictive AI models, yet internal oversight for these tools remains remarkably inconsistent across the industry. National data indicates that among hospitals utilizing predictive models, only about 61% conduct accuracy testing using local patient data, while a mere 44% assess these models for algorithmic bias before implementation. Despite these significant gaps in quality control, many institutions continue to argue that because these models are hosted and used internally to care for their own patient populations, they should remain exempt from formal FDA oversight. This perspective classifies AI tools as a form of professional self-regulation and institutional governance rather than as commercial medical products. However, such a decentralized approach allows for wide variations in performance, potentially exposing patients to risks that would be unacceptable in any other area of medical technology today.
The Impact of a Two-Tier Regulatory System
The absence of a unified federal standard has resulted in what Gottlieb describes as a “two-tier” market, creating an uneven playing field that fundamentally disadvantages commercial developers. On one side, academic centers can rapidly iterate and deploy high-value algorithms with minimal regulatory burden, benefiting from the agility of internal development cycles. On the other side, commercial entities aiming to bring similar or even identical tools to the broader market must navigate the rigorous, time-consuming, and expensive FDA medical device review process. This disparity not only creates significant economic imbalances but also stifles the healthy competition necessary for long-term technological advancement. Furthermore, it ensures that a substantial portion of the clinical AI landscape exists without federal safety benchmarks, leaving clinicians and patients without the assurance that their digital health tools have met a universal standard of excellence across various clinical settings.
The Proposed Certification Model: A New Path for Health Systems
Adapting the CLIA Framework for Artificial Intelligence
To bridge this regulatory gap, the proposed framework draws inspiration from the Clinical Laboratory Improvement Amendments, which historically managed the oversight of laboratory-developed tests. Historically, these tests existed in a similar regulatory vacuum until the CLIA framework provided a way for the Centers for Medicare and Medicaid Services to ensure quality without stifling the localized needs of specialized labs. Under a “CLIA-like” system for clinical AI, the regulatory focus would shift from the specific algorithm to the institution itself. Instead of reviewing every individual update to a piece of software, the FDA would certify health systems based on their proven ability to meet strict standards for validation methodologies, training data diversity, and intended clinical use. This approach acknowledges that high-performing health systems often have the technical expertise to manage their own digital assets, provided they operate under a set of federally mandated safety guidelines and protocols.
Future Implications: Standardizing Excellence in Clinical AI
The transition toward a centralized oversight model for local governance provided a clear solution for the challenges of clinical AI integration. Stakeholders realized that by formalizing the validation process, they removed the primary barriers to cross-institutional collaboration and commercial scaling. The adoption of these standards empowered academic centers to share their most effective predictive tools with smaller community hospitals, effectively democratizing high-quality digital care. Moving forward, the industry prioritized the development of interoperable data standards and bias-mitigation protocols that became the bedrock of the certification process. Legislative leaders recognized the urgency of the situation and began drafting the necessary amendments to clarify the FDA’s role in this new ecosystem. This structural change shifted the focus from reactive risk management to proactive quality assurance, ensuring that every patient benefited from the same level of algorithmic safety in modern medicine.
