Artificial intelligence is moving from laboratory demonstrations into clinical use, with several technologies already helping patients while others advance through trials. Researchers are using AI to personalize cancer treatments, recover hidden reproductive cells, discover overlooked uses for existing drugs, and flag cardiovascular disease years before symptoms appear. However, the gap between promising research and available treatment remains significant, and patients deserve clear answers about how AI affects their care.
The most recent milestone involves a personalized melanoma treatment. On August 19, Moderna and Merck announced positive results from a Phase 3 trial testing intismeran autogene, also known as V940 or mRNA-4157, combined with Keytruda. The trial enrolled 1,137 people with high-risk melanoma that surgeons had completely removed. The combination met its primary endpoint for recurrence-free survival and a key secondary endpoint measuring distant metastasis-free survival. The companies called this the first positive Phase 3 readout for an individualized neoantigen therapy and the first positive Phase 3 result for an mRNA-based cancer therapy.
The treatment works by analyzing each patient’s tumor mutations. Researchers extract a tumor sample, identify its unique mutations using an algorithm, and select targets that may help the immune system recognize the cancer. The resulting individualized therapy can encode up to 34 neoantigens. Moderna has stated the V940 program uses integrated AI algorithms during development. The company then creates an mRNA treatment based on the selected targets. In a smaller Phase 2b study with longer follow-up, intismeran plus Keytruda reduced the risk of recurrence or death by 49 percent compared with Keytruda alone and reduced the risk of distant metastasis or death by 59 percent.
The melanoma results are encouraging, but important context applies. Merck and Moderna have announced only topline results from the Phase 3 trial so far. The companies plan to present full findings at an international medical meeting and share them with regulators. The study continues to track overall survival. Intismeran remains investigational, and the FDA has not approved it as a melanoma treatment.
Using AI to Find Treatments Already in Use
While Moderna and Merck develop new therapies, another group is asking whether useful treatments already exist. Dr. David Fajgenbaum co-founded the nonprofit Every Cure to pursue that possibility. According to Every Cure’s 2025 annual report, about 18,000 recognized diseases exist worldwide, but roughly 4,000 have FDA-approved medications. That leaves an enormous number of diseases with limited treatment options.
Every Cure uses AI to scan biomedical knowledge and look for connections between existing medicines and other diseases they might potentially treat. The organization says its system can generate tens of millions of predictions in less than a day. Researchers then examine the most promising possibilities. The federal Advanced Research Projects Agency for Health, known as ARPA-H, is backing this approach through a project called MATRIX, which uses machine learning and artificial intelligence to predict which FDA-approved drugs could potentially treat other diseases.
A real case shows why this matters. Kaila Mabus developed multicentric Castleman disease at 13 and became severely ill despite chemotherapy. In 2020, her doctors tried ruxolitinib, a drug already used for certain blood disorders but not FDA-approved for Castleman disease. She began improving within months and was declared in remission in January 2021. AI did not identify her treatment, but her case demonstrates why Every Cure wants to use AI to uncover promising drug-disease connections much faster and on a far larger scale.
Finding What Human Eyes Miss
At Columbia University Fertility Center, artificial intelligence has taken on a different challenge. Researchers developed the Sperm Tracking and Recovery system, known as STAR. It combines high-speed imaging with an AI detection model and microfluidics. STAR was designed for patients with azoospermia or cryptozoospermia, conditions where sperm may appear absent or exist in extremely small numbers.
The system examines a semen sample far more thoroughly than a person could do by hand. STAR can capture and process about 1.1 million images every hour. Its AI model examines frames for possible sperm cells. When the system confirms one, a microfluidic mechanism isolates the cell. In one validation sample, embryologists searched for two days without finding sperm. STAR found 44 sperm in about an hour.
STAR achieved its first reported pregnancy in March 2025. The couple involved had spent nearly two decades trying to conceive. STAR found and recovered sperm that conventional examination of the same sample had missed. The pregnancy later resulted in a healthy delivery. However, this technology does not work uniformly. Columbia reports that sperm are found in about 28 percent of patients who previously received an azoospermia diagnosis. About 20 percent of mature eggs fertilize with STAR-recovered sperm, and around 18 percent of those fertilized eggs develop into good-quality embryos for transfer or freezing. Those rates are lower than typical IVF outcomes, but the patients using STAR often face especially difficult fertility problems.
Earlier Warning for Heart Disease
Researchers at the University of Hong Kong are exploring AI’s potential for cardiovascular risk. Their AI-based tool, called CardiOmicScore, analyzes molecular information found in blood. The system examined 2,920 circulating proteins and 168 metabolites, and incorporated genomic information from data in the UK Biobank.
CardiOmicScore uses deep learning to estimate future risk of six cardiovascular diseases: coronary artery disease, stroke, heart failure, atrial fibrillation, peripheral artery disease, and venous thromboembolism. Researchers found that the approach improved risk prediction when combined with standard clinical information. In some cases, CardiOmicScore could flag elevated risk as much as 15 years before symptoms appeared. Instead of discovering cardiovascular disease after symptoms develop, doctors might get a warning while there is still time to intervene. However, CardiOmicScore remains a research development. You cannot walk into your doctor’s office today and request it as a routine screening test.
What Patients Should Ask
You may encounter AI in healthcare without ever opening an AI chatbot. A laboratory could use it while analyzing a tumor. A fertility clinic might use it to search for something the human eye missed. Researchers can also use AI behind the scenes to find treatments worth investigating. The key question is how much evidence supports the specific technology being used. A university research project sits at a very different stage from a medical device that has gone through clinical testing and regulatory review.
When AI becomes part of your care, ask four simple questions. First, find out what the AI actually does and what role the technology plays in your care. Second, ask whether a doctor, specialist, or laboratory professional checks the AI’s findings before anyone makes a decision. Human review becomes especially important when a result could affect treatment or diagnosis. Third, ask whether the FDA has cleared or approved the technology when regulatory authorization applies, and what type of research supports it. Finally, ask how your provider stores your health data and who can access it.
AI can help doctors process information and identify patterns that would be difficult to find alone. Your healthcare decisions still deserve qualified medical judgment based on your individual circumstances. The science continues to advance, but the evidence for some technologies remains limited. What matters most is understanding whether the specific AI your doctor is using has proven itself in the settings where you need it.
