Clinical Harness: A Governance Architecture for Medical AI Skills.

The paper introduces a novel architectural framework called the "Clinical Harness," aimed at addressing the current fragmentation of AI in medical settings. Instead of relying on isolated AI models, this system envisions AI as a collection of accountable, persistent "skills" that can be governed and orchestrated throughout a patient's care journey. The Clinical Harness provides a robust runtime governance layer for medical AI. Its functions include registering new AI capabilities, orchestrating their use in complex clinical workflows, guarding against misuse or errors, and continuously monitoring their performance and impact. This ensures that AI applications in healthcare are not only effective but also safe and reliable. Using osteoporosis as a practical example, the authors illustrate how diverse AI skills—ranging from knowledge-driven to data-driven and physics-enhanced approaches—can be integrated and managed under this governance architecture. This allows for comprehensive, lifecycle-oriented care, ensuring that AI supports patients consistently over time.