Change Management for Generative AI-Enabled Medical Devices: Adapting the Predetermined Change Control Plan Framework

GenAI-enabled medical devices combine foundation models, prompts, retrieval sources, tools, orchestration, guardrails, and output processing, which jointly determine clinical behavior. Changes to these devices may span several interacting components, occur without a discrete release, or originate from a third-party foundation-model developer. FDA-authorized Predetermined Change Control Plans (PCCPs) prospectively specify allowable device modifications and their validation and assessment. To determine how PCCPs have been used in GenAI-enabled medical devices, we analyzed 3,758 FDA 510(k) decisions from May 31, 2025 through August 10, 2026. We identified 106 records with authorized PCCPs, including at least 37 clearances for AI/ML-enabled devices. This showed that the FDA has authorized PCCPs to support model retraining, input expansion, pipeline modification, and bounded architecture changes. We found no clear examples covering the GenAI stack, which is consistent with the limited number of GenAI-enabled devices that have been authorized. We propose that a PCCP should permit foundation-model replacement and related changes to prompts, retrieval, tools, and guardrails without prespecifying the exact successor configuration. This flexibility is necessary because rapidly evolving foundation models may require unanticipated, interdependent changes across the GenAI stack. Similar to PCCP-eligible changes for other AI-enabled devices, an impact assessment should determine the appropriate validation, monitoring, and rollback controls.