AI-discovered drugs still have to succeed in clinical trials the same way every other drug does: by demonstrating safety and efficacy in people. Computational design may improve a candidate's odds or speed its arrival, but it does not exempt it from human biology, regulatory standards or the operational realities of running trials.
The first wave of AI-designed molecules has now moved through early clinical studies. One widely followed example is Insilico Medicine's candidate for idiopathic pulmonary fibrosis, which reported Phase 2a results in 2025 after target discovery and molecule design were driven by AI. Examples like this are encouraging. They are also a reminder that the clinical path remains long.
Do AI-discovered drugs succeed more often in trials?
Early analyses have suggested that AI-discovered molecules may show higher success rates in Phase 1, where safety and pharmacokinetics are assessed. That would be consistent with AI's strength in optimizing drug-like properties. The evidence for later phases, where efficacy must be proven, is still limited. Efficacy depends on whether the biological hypothesis is right, and that is exactly what clinical trials exist to test.
In other words, AI may be better at producing a good molecule. Whether that molecule treats the disease is still an empirical question.
Regulators evaluate the drug, not the method
Regulators assess a candidate on its data: nonclinical studies, manufacturing quality and clinical evidence. How the molecule was discovered does not lower the bar. Where AI is used to generate evidence submitted to regulators, such as modeling that informs dose selection, FDA's January 2025 draft guidance suggests sponsors should be prepared to demonstrate the credibility of that model for its context of use.
Operational realities do not change
An AI-designed drug still needs sites, investigators, participants, monitoring, data management and safety oversight. It still faces the same enrollment challenges and the same risk of protocol amendments. For novel targets, operational challenges may be greater, because there are fewer precedent trials and less established site experience.
This is where experienced clinical development judgment adds value. Designing a first-in-human study for a novel mechanism, choosing sites with relevant expertise, and planning adaptive elements responsibly all depend on people who have seen trials succeed and fail.
What sponsors should plan for
Treat clinical development as the main determinant of timeline, not discovery speed.
Build flexibility into early-phase design, such as adaptive dose escalation where appropriate.
Prepare to explain any AI-derived evidence used in regulatory submissions.
Invest in biomarker strategy to learn as much as possible from each study.
AI is changing how drugs are found. The clinic is still where we learn whether they work. That is not a limitation of AI. It is the point of clinical research.
