Researchers at the École Polytechnique Fédérale de Lausanne have launched MeditronFO, a fully open framework designed to build and audit clinical large language models. Developed by the Laboratory for Intelligent Global Health & Humanitarian Response Technologies, the initiative addresses growing concerns over proprietary medical AI systems that conceal their training data, design choices, and decision-making processes. MeditronFO establishes a transparent pipeline from dataset curation to model evaluation, enabling independent scrutiny by clinicians, hospitals, and regulators. Building upon the 2023 Meditron project, the new framework adapts open base models such as OLMo, EuroLLM, and Apertus, a Switzerland-based model developed jointly by EPFL and ETH Zurich. Unlike partially open systems that release only the final weights, MeditronFO publishes all underlying datasets, processing methodologies, training code, and evaluation metrics. The team constructed the Meditron Corpus by integrating publicly available medical literature with clinician-validated synthetic data derived from clinical guidelines and realistic patient scenarios. More than 46,000 expert-curated practice guidelines informed the dataset, ensuring alignment with real-world medical standards. Clinician involvement remains central to the development lifecycle. Through the MOOVE validation platform, medical professionals actively audit training materials, validate model outputs, and identify safety risks throughout the refinement process. This participatory approach aims to prevent reliance on external proprietary platforms whose development priorities may not align with local healthcare needs. Initial testing demonstrates that full transparency does not compromise performance. Every variant tested outperformed its original base model, with Apertus-70B-MeditronFO recording a 6.6 percentage point improvement on standardized medical examinations. Researchers published their methodology and benchmarks on the arXiv preprint server, emphasizing that auditable pipelines can produce competitive medical AI systems without sacrificing accountability. The release marks a strategic shift in health technology governance. EPFL and the LiGHT lab are now preparing the MED.USE initiative, a multi-year clinical trial spanning facilities in Switzerland and Tanzania. The study will monitor how practitioners interact with AI-generated recommendations, assess impact on patient outcomes, and measure whether transparency-driven tools reduce unnecessary interventions. Lead researcher Xavier Theimer-Lienhard noted that the medical field operates on verifiable standards, a principle now extended to AI development. Project director Professor Mary-Anne Hartley stressed that the medical AI ecosystem must prioritize scientific scrutiny and clinician participation over black-box commercial solutions. By proving that openness and clinical performance can coexist, MeditronFO offers a scalable alternative to closed-source health AI. The framework establishes a new benchmark for regulatory compliance, institutional data sovereignty, and collaborative model development, positioning transparency as a competitive advantage rather than a technical limitation.