Artificial intelligence is beginning to reshape the way pharmaceutical companies evaluate experimental medicines, with AI virtual drug trials emerging as a potential tool for identifying clinical risks before costly studies involving real patients begin. The technology is being explored by drugmakers and specialized AI companies as the industry looks for ways to reduce the number of promising treatments that ultimately fail during human testing.
The need for better prediction is significant. Developing a new medicine can take years and require enormous financial investment, yet only a relatively small share of drug candidates eventually receives regulatory approval. Pharmaceutical companies must move experimental treatments through several stages of clinical research, beginning with early studies focused largely on safety and progressing to larger trials that determine whether a treatment actually provides meaningful benefits to patients.
AI companies believe virtual trials could help address one of the industry’s biggest challenges: discovering potential problems before a medicine reaches an expensive late-stage study. Instead of relying solely on traditional clinical testing, researchers can use computer models to simulate how groups of virtual patients might respond to a treatment. These simulations can examine different patient characteristics, treatment responses and potential outcomes within a much shorter period.
The approach gained attention after an AI startup called BioinvestGPT conducted a simulated trial involving Novartis’ experimental drug del-desiran, which was being developed for a rare form of muscular dystrophy. Before a late-stage clinical trial produced disappointing results, the AI company’s simulation had predicted that the treatment would deliver only a limited clinical benefit.
The subsequent trial failure highlighted why pharmaceutical companies are increasingly interested in predictive technologies. Novartis had previously viewed del-desiran as a potentially important commercial product, with expectations that it could eventually generate billions of dollars in annual peak sales. When the late-stage study failed to meet its objective, the company’s shares fell sharply, illustrating the enormous financial consequences that can follow an unsuccessful clinical program.
AI simulations are not intended to replace human clinical trials. Instead, companies developing these systems see them as an additional layer of analysis that can help determine whether a drug program deserves to proceed to the next stage. A simulation may allow researchers to test assumptions about a treatment, identify populations that could respond differently, or uncover warning signs that might otherwise emerge only after a trial has already begun.
Francisco Beca, chief medical officer at QuantHealth, an AI clinical trial simulation platform based in Tel Aviv, described the broader objective in simple terms. “We shouldn’t only ask how to run trials faster. We should ask how to run fewer trials that are going to fail,” he said.
That distinction is important because speeding up clinical development alone does not solve the industry’s fundamental problem. A faster trial that produces an unsuccessful result still consumes significant resources and exposes patients to an experimental treatment that may ultimately prove ineffective. The greater value of AI could therefore come from improving decisions about which treatments should enter human testing in the first place.
Traditional clinical development follows a structured sequence. Phase 1 studies generally involve relatively small groups and focus primarily on safety and appropriate dosing. Phase 2 trials provide more information about whether a treatment appears to work and help researchers understand its risks. Phase 3 studies are substantially larger and provide the evidence regulators typically require when evaluating a medicine’s effectiveness and safety.
Moving through these stages can take several years. AI simulations, by comparison, can sometimes produce results in weeks or even less than a month. This difference in speed gives researchers an opportunity to explore multiple scenarios before committing the considerable time and money required for a conventional clinical trial.
The potential financial impact is also attracting attention. The global biopharmaceutical industry spends an estimated $140 billion each year on human clinical testing, while only about 12% of drug candidates eventually receive regulatory approval. The relatively low success rate has remained a persistent challenge for pharmaceutical companies despite decades of improvements in biomedical research and clinical trial design.
AI companies argue that better forecasting could improve those numbers by helping drug developers eliminate weaker candidates earlier. The technology can also be used when pharmaceutical companies evaluate potential acquisitions or licensing opportunities. Rather than judging a drug candidate solely on laboratory results and existing clinical data, companies can use simulations to estimate how it might perform under different clinical conditions.
The growing investment in AI drug development reflects confidence that these tools could become an important part of pharmaceutical research. Funding for AI-driven drug discovery more than doubled between 2023 and 2025, reaching billions of dollars. Much of that investment has concentrated on areas where AI has already demonstrated practical value, including molecular design and the search for potential drug candidates.
Clinical development remains a more complicated challenge. Predicting how a molecule will behave is only one part of determining whether a medicine will help people. Human biology is highly variable, and factors such as age, genetics, existing medical conditions, medication use and adherence can influence treatment outcomes. Even sophisticated computer models cannot completely reproduce the complexity of a living human body.
This creates an important limitation for virtual clinical trials. AI simulations depend heavily on the quality of the data used to build them. If the underlying information is incomplete, biased or poorly representative of the population that will eventually receive the medicine, the predictions may also be unreliable. A model that performs well in one group of patients may not necessarily produce equally accurate results in another.
There is also the question of how regulators will ultimately assess AI-generated evidence. Traditional clinical trials operate under established standards designed to protect participants and determine whether medicines are safe and effective. Virtual simulations may become increasingly influential in planning these studies, but they are unlikely to remove the need for carefully controlled human research.
The most realistic role for AI may therefore be as a decision-making tool rather than a replacement for clinical trials. Pharmaceutical companies could use simulations to identify promising candidates, refine trial designs, select appropriate patient populations and recognize possible risks before exposing large numbers of people to an experimental treatment.
