Medical artificial intelligence is bifurcating into two distinct tracks: research models that post strong benchmark results but lack clinical authorisation, and workflow tools already embedded in health systems at scale.
Microsoft's Nuance DAX Copilot, an AI clinical documentation assistant, now operates in more than 150 health systems, while research from Kaiser Permanente found AI scribes saved physicians an estimated 15,791 hours.
Specialised speech models such as Nova-3 Medical, built for healthcare settings, have reached a 3.44% word error rate and are being deployed in environments compliant with the Health Insurance Portability and Accountability Act (HIPAA), the US federal standard governing patient data privacy.
In drug discovery, Insilico Medicine reported positive Phase IIa trial results for ISM001-055, its AI-discovered treatment for pulmonary fibrosis, a scarring of lung tissue.
The highest-dose patient group improved forced vital capacity, a measure of lung function, by 98.4 millilitres over 12 weeks, against a 62.3 millilitre decline for the placebo group, a differential of roughly 160 millilitres.
Insilico says its platform reduced development time to preclinical candidate by more than 60%.
AlphaFold 3, the protein-structure prediction system developed by Google DeepMind that earned a Nobel Prize, uses a diffusion-based architecture to model interactions between proteins, DNA, RNA and small molecules, and underpins a public database of 214 million predicted protein structures.
Related reading
- OpenAI publishes open-source teen safety tools for developers building AI apps
- Microsoft and Nvidia team up to bring AI to nuclear power permitting and design
- Microsoft tightens Edge for Business controls to block unauthorised AI tools
The US Food and Drug Administration has issued a framework titled "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products," establishing risk-based credibility assessments across development stages.
Analysts expect the first regulatory approvals of fully AI-discovered drugs within the coming years, as clinical validation and regulatory review continue.
The recap
- Medical AI splits between benchmark research and deployed workflow tools.
- AlphaFold database contains 214 million predicted protein structures.
- FDA expects first AI-discovered drug approvals within coming years.
