How AI Is Changing Diagnosis, Drug Discovery and Patient Care in 2026

Artificial intelligence is now built into more than 1,600 medical devices authorized for the U.S. market, and it has expanded from reading scans to designing entirely new drug molecules. In July 2026, Insilico Medicine advanced rentosertib — a compound whose biological target and chemical structure were both generated by AI software — into a 320-patient Phase III trial for idiopathic pulmonary fibrosis, a lung disease with few treatment options. At the same time, global health bodies are still working out how to govern systems that increasingly shape diagnosis and treatment decisions.

Where AI Already Works: Reading Images

The U.S. Food and Drug Administration (FDA) states it has authorized more than 1,600 AI-enabled medical devices for marketing in the United States, according to the agency's own tracking page, last updated September 22, 2026. Examples the FDA lists include software that screens skin lesions for cancer, detects diabetic retinopathy from retinal photographs, estimates heart-attack probability from sensor data, and automates insulin dosing based on glucose-monitor readings.

An analysis of 2025 FDA 510(k) clearances by regulatory consultancy Innolitics found that radiology accounted for the large majority of that year's AI/ML device authorizations, with 295 total clearances issued to 221 different manufacturers and a median review time of 142 days. The single most common category was radiological computer-aided detection software, which alone made up more than a quarter of all 2025 clearances, according to the analysis.

Cardiology, neurology and orthopedics trailed far behind radiology in 2025, and specialties such as anesthesiology and clinical chemistry each accounted for under 2 percent of clearances, per the same analysis.

AI Is Designing Drugs, Not Just Screening Them

Beyond diagnostics, AI software is now being used to propose new drug targets and design the molecules meant to hit them, a step beyond earlier uses of computing to simply screen existing compound libraries faster. Insilico Medicine, a Hong Kong-based biotechnology company, says its PandaOmics platform identified a protein called TNIK as a disease target and its Chemistry42 platform generated a candidate molecule, now named rentosertib, for idiopathic pulmonary fibrosis. The company says the discovery pathway was described in a paper in Nature Biotechnology.

In a randomized, placebo-controlled Phase IIa trial, patients taking the highest tested dose of rentosertib (60 mg once daily) saw lung function, measured as forced vital capacity, improve by a mean of 98.4 milliliters over 12 weeks, compared with a mean decline of 20.3 milliliters in the placebo group, according to Insilico Medicine's own announcement of the results, which the company says were also published in Nature Medicine.

On July 7, 2026, Insilico said it began a Phase III trial of rentosertib: a randomized, double-blind, placebo-controlled study enrolling 320 patients across 47 centers in China over 52 weeks. The primary endpoint is the annual rate of decline in forced vital capacity, with time to first disease-progression event as a key secondary measure, the company said.

No AI-designed drug has yet received approval from the FDA or any other regulator, and rentosertib's lung-function improvements in the Phase IIa trial were listed as secondary, exploratory endpoints rather than the study's primary safety goal, by the company's own description. Industry trackers cited by the analysis site IntuitionLabs counted more than 170 AI-discovered drug programs in clinical development in early 2026, up from roughly two dozen in late 2023, though IntuitionLabs noted the figures vary between trackers because no single authoritative, government-maintained registry of such drugs exists and each tracker applies its own classification criteria.

Regulators Are Still Writing the Rules

The World Health Organization (WHO) issued ethics and governance guidance for large multi-modal AI models in health on January 18, 2024, setting out more than 40 recommendations for governments, developers and health workers. The guidance states that poor-quality or biased training data can produce inaccurate or incomplete health information, and that clinicians risk falling into what it calls 'automation bias' by over-trusting AI outputs or improperly delegating decisions to them.

• Biased or low-quality training data can skew outputs by race, ethnicity, sex, gender identity or age
• Automation bias: health workers may overlook AI errors or delegate decisions inappropriately
• Cybersecurity gaps could expose patient data or undermine trust in AI systems
• High-performing AI tools may remain accessible mainly to wealthier health systems, widening global gaps in care

WHO Chief Scientist Jeremy Farrar said at the time of the guidance's release that generative AI technologies could improve health care only if developers, regulators and users identified and fully accounted for the associated risks.

What Hasn't Happened Yet

Several milestones sometimes associated with AI in medicine have not occurred. No fully AI-designed drug has secured regulatory approval anywhere, and WHO's guidance explicitly warns that unequal access to the most capable AI systems could widen, rather than narrow, disparities between well-resourced and under-resourced health systems. For now, AI's clearest footprint in medicine remains diagnostic imaging, where regulators have the longest clearance history, while its role in creating new medicines is moving from laboratory proof-of-concept toward larger human trials.