An evidence informed framework for artificial intelligence in rare breast cancers using small cohort validation synthetic data practices and clinical governance.

Rare breast cancers represent a clinically important but underrepresented group of malignancies. In this Perspective, rare breast cancers are considered within the broader rare cancer definition of an annual incidence below 6 cases per 100,000 persons, while also recognizing breast-specific rarity based on uncommon histology, molecular hallmarks, clinical presentation or sex-specific occurrence. These conditions are characterized by limited case numbers, biological heterogeneity, reduced clinical trial inclusion and fragmented evidence. These constraints challenge artificial intelligence (AI) development because many systems depend on large, balanced and externally validated datasets. AI may support diagnosis, histopathology, molecular interpretation, prognostic stratification and precision oncology decision support, but its use in rare breast cancers requires evidence standards adapted to small cohorts. This Perspective proposes an evidence-informed clinical governance framework organized around five domains: intended clinical use, small-cohort validation, synthetic data governance, human oversight and lifecycle monitoring. Its distinctive contribution is to translate general AI reporting and governance principles into rare breast cancer-specific safeguards, including objective data-quality checks, leakage prevention, uncertainty-aware validation, synthetic data plausibility scoring, pan-rare model reporting, patient involvement and post-deployment surveillance. Synthetic data may support development and simulation, but should not replace validation on real clinical cases. By linking small-cohort methodology with clinical oversight, regulatory alignment and lifecycle monitoring, the framework offers a practical roadmap for safe AI-enabled rare breast cancer precision oncology.