Women's-health AI/ML tooling

Research on female populations becomes possible at scale when computational infrastructure is built for it: open-source models trained on women’s health data, bias-audited against sex and gender, with regulatory guidance holding developers to that standard. The regulatory clocks are running. EU AI Act Article 10 mandates bias examination and mitigation in training data for high-risk AI systems. Medicaldevice high-risk obligations under Annex I and Article 6(1) apply from 2 August 2027, while Article 57’s regulatory sandbox availability deadline is the nearer 2 August 2026.

Nordic-scale compute and clinical leadership are now in place. DTU and Amager-Hvidovre Hospital received access to Denmark’s Gefion supercomputer at the Danish Centre for AI Innovation (DCAI) on 4 September 2025, with women’s health AI projects in the early stages. What is missing is a shared open-source toolkit, bias-audited against an agreed consensus standard. Without one, every developer either builds the audit alone or skips it, and Nordic women’s health AI flows offshore. Oura’s redomiciliation to Delaware in February 2026 is the worked example.

A Nordic open-source toolkit hosted on existing infrastructure (Gefion, NeIC, LUMI), bias-audited against the STANDING Together consensus framework, would build on the Hvidovre, DTU, and DCAI foundation and the wider Nordic clinical AI ecosystem rather than create a parallel one. The result is that a Stockholm hospital, a Bergen research group, and a Copenhagen startup audit against the same standard, and the regulatory cost of doing women’s health AI research in the Nordics drops. DTU, Hvidovre, and DCAI are positioned to lead on the operational side, with AI Sweden, Karolinska’s WASP and WARA Medicine, the Datatilsynet sandboxes in Norway and Denmark, and the NeIC Nordic AI Union pre-study leads as the assembly partners.