Health & medicine
Diagnosis, admin load, clinical research — AI lands on all three fronts.
Healthcare AI is no longer experimental. By 2026, three clinically adopted use cases stand out: diagnostic support, admin automation and care copilots. Regulation remains heavy — the AI Act treats them as high-risk by default.
Family of dedicated medical models. 86.5% accuracy on USMLE questions — +19% versus its predecessor. To deploy via Vertex AI for internal clinical apps.
LLM dedicated exclusively to NON-diagnostic tasks: patient education, adherence reminders, pre-visit prep, post-discharge follow-up. Built-in medical safeguards.
Differential diagnosis support and clinical plan drafting for the physician. Reference in the US, rolling out at a few French university hospitals.
Pathology slide analysis. Assisted oncological diagnosis, tumour characterisation, therapeutic research.
None of these tools replaces medical decision-making. But MedLM via Vertex (to build your own apps) and Hippocratic (to absorb administrative load) are the two bets to investigate if you run a care organisation.