Recent research published in the Journal of Medical Internet Research highlights how machine learning is helping to bridge critical evidence gaps regarding drug safety for pregnant women. A news and perspectives article titled How Machine Learning Can Help Close Evidence Gaps for Drug Safety in Pregnant Women features interviews with principal investigators behind two major initiatives designed to analyze vast datasets of medication exposure and outcomes to identify potential risks. The study addresses a longstanding crisis in medical research: the severe underrepresentation of pregnant women in clinical trials. According to health writer Michelle Falci, who conducted the interviews, only 4% of clinical trials over the last decade included pregnant participants. This exclusionary trend stems from a 1977 recommendation by the US Food and Drug Administration, which advised against including pregnant women or those capable of becoming pregnant in early-phase clinical trials. While intended to protect fetuses, this policy created a decades-long scarcity of data on medication safety for expectant mothers, contributing to broader underrepresentation of women in medical studies. Despite various efforts to fill this void, practical outcomes have remained insufficient. Two new projects are leveraging artificial intelligence to address this shortfall. The first, known as the BOOST-HP project, utilizes a tree-based approach to data mining. The second, the BIONIC study, combines causal inference techniques with machine learning. Both approaches rely on advanced algorithms to process large-scale data, enabling researchers to monitor exposure patterns and estimate potential causal links between specific medications and adverse pregnancy outcomes. Experts emphasize that while the technology holds immense promise, it requires careful implementation. Cristina Longo, the leader of the BIONIC study, notes that these AI-assisted methods will benefit significantly from access to more comprehensive datasets. Almut G. Winterstein, a principal researcher on the BOOST-HP project, stresses the importance of transparency. She and her team utilize AI models capable of tracing the decision pathways that lead to their evaluations. Winterstein warns against using so-called black box models, where the internal logic is opaque, as this could obscure crucial epidemiological errors. Instead, the researchers advocate for thoughtful model design and a commitment to explainable AI. By ensuring that the reasoning behind AI predictions is clear and auditable, the scientific community can better trust and apply these findings to clinical practice. Although challenges remain regarding data availability, the convergence of machine learning and epidemiology offers a powerful tool to finally generate the safety evidence needed for pregnant women. This shift marks a potential turning point in ensuring that medication guidelines are based on robust data rather than assumptions born from historical exclusion.