Why Central AI Governance Committees Are Failing Healthcare—And Their Fix
If health systems, payers and pharma companies want to move from dozens of AI pilots to hundreds of production systems, the manual committee model has to change.
David Talby, Forbes Councils Member
Forbes
2 min read
6/10
Key Takeaways
Fewer than 5% of healthcare AI pilots reach full production, largely due to central committee bottlenecks that take 6–12 months per model review.
The FDA cleared over 1,000 AI/ML medical devices by 2025, placing unprecedented strain on manual governance committees in large hospital systems.
Decentralized governance with automated guardrails can cut AI model review time by up to 80%, as demonstrated by early pilot programs at Intermountain Healthcare.
Healthcare organizations using federated governance models report 3× faster AI deployment and a 40% reduction in compliance incidents compared to centralized committees.
Low-risk AI applications (e.g., scheduling, billing) can be approved through self-service dashboards, freeing committees to focus on high-risk clinical algorithms.
Central AI governance committees, once hailed as the guardians of safe and ethical AI in healthcare, are now the primary bottleneck preventing the industry from scaling artificial intelligence. These manual review boards are failing health systems, payers, and pharmaceutical companies that need to move from dozens of pilots to hundreds of production AI systems—and a radical fix is required. Healthcare organizations have invested billions in AI, yet most models never leave the pilot stage. The culprit, according to experts cited in a new Forbes analysis, is the centralized committee model that approves each AI application one at a time. These groups—often composed of clinicians, data scientists, legal teams, and compliance officers—meet weekly or monthly, reviewing each model manually against a checklist of safety, bias, and efficacy criteria. With the average health system running 10 to 50 AI pilots simultaneously, the backlog can stretch to months or years. The problem is now acute: the FDA cleared over 1,000 AI-enabled medical devices by 2025, and hospital systems are drowning in governance requests. The fix, the article argues, is a decentralized, automated governance framework. Instead of a single committee, organizations should embed governance into the AI development lifecycle—using automated guardrails, continuous monitoring, and tiered approval processes. For example, low-risk administrative AI could be approved via self-service dashboards, while high-risk clinical algorithms undergo rigorous automated testing and human review only when breaches occur. Early adopters like Intermountain Healthcare and Geisinger are already piloting such systems, reporting 80% faster time-to-production and fewer compliance incidents. The shift mirrors what cybersecurity did decades ago: moving from manual perimeter checks to embedded, continuous monitoring. The broader implication is that healthcare AI has reached an inflection point. If governance does not evolve, the industry risks either stifling innovation or deploying unsafe AI. Industry observers predict that within three years, most leading health systems will adopt some form of federated governance, combining automated checks with sparse human oversight. The milestone to watch is the first hospital that achieves a 90% AI pilot-to-production conversion rate—a target that seems impossible today but is within reach if governance committees stop being the gatekeepers and start being enablers.
Frequently Asked Questions
Central committees fail because they use manual review processes that cannot keep pace with the number of AI pilots. Health systems running 10–50 pilots see approval backlogs of 6–12 months, causing most models to never reach production.
The fix is a decentralized, automated governance framework that embeds controls into the AI development lifecycle. Automated guardrails, tiered approval levels, and continuous monitoring replace slow manual committee reviews.
Organizations can scale by separating low-risk AI (approved via self-service dashboards) from high-risk AI (subject to automated testing and limited human review). This approach cuts review time by up to 80% and increases deployment rates.
Alternatives include federated governance models where authority is distributed to domain-specific teams, and AI-driven governance platforms that use algorithms to monitor compliance and flag issues automatically.
Under central committees, AI model review typically takes 6–12 months. With automated governance, the review time can drop to a few weeks or even days for low-risk applications.
