Background: Natural history collections are foundational infrastructure for biodiversity science, but the global geography of specimen custody — where specimens are housed relative to where they were collected — has rarely been quantified through time. Understanding these patterns is essential for assessing equitable access, providing information for capacity-building efforts and ensuring that countries of origin benefit from biodiversity research.

Objectives: We provide the first time-resolved, country-to-country analysis of specimen flow networks for vertebrates, using three globally distributed mammalian families (Canidae, Felidae, Mustelidae) as case studies. We ask: (1) How have regional shares of specimen holdings changed over time? (2) Have holdings become more evenly distributed across countries? (3) How have origin→holding flows shifted with respect to regional retention and reciprocity? (4) Do the three families exhibit distinct curation profiles?

Methods: We analysed 253,131 preserved-specimen records from GBIF spanning 1900–2020, aggregated into four approximately equal temporal periods (1900-1929, 1930-1959, 1960-1989, 1990-2020). We constructed directed, weighted country-to-country flow networks for each family × time period combination and computed network metrics (size, complexity, density, reciprocity, connectivity) and inequality metrics (Gini coefficient). We visualised geographic patterns using network maps on Robinson projections and assessed temporal trends using statistical models.

Results: Specimen holdings are highly concentrated: the United States holds 55% of all specimens and the top 10 countries account for 90%. Inequality follows a U-shaped trajectory: high concentration in early periods (Gini 0.85–0.89 in 1900-1929), democratisation by the 1960-1989 period (Gini 0.77–0.89), then re-concentration in 1990-2020 (Gini 0.81–0.86). Specimen flow networks underwent simplification, contracting from 100–111 participating countries and 207–262 flows in the 1930-1959 period to only 75–84 countries and 115–160 flows in 1990-2020 — a 23–29% reduction in network size and 39–56% reduction in complexity. Despite simplification, networks remain dense and cohesive amongst remaining participants. However, reciprocity (bidirectional exchange) is low (mean 0.075 or 7.5% in recent periods), indicating predominantly unidirectional flows from biodiversity-rich regions to museum-rich nations. These patterns are broadly consistent across Canidae, Felidae and Mustelidae, though Felidae exhibits less inequality rebound, possibly reflecting conservation-driven capacity building.

Conclusions: Our findings reveal persistent geographic imbalances in specimen custody, characterised by concentration in wealthy nations, low reciprocity and network contraction over time. While some patterns may reflect digitisation lags, genuine reductions in the breadth of global collecting and shifts towards extractive (non-reciprocal) flow structures are evident. These imbalances create structural inequities in research access, taxonomic expertise and capacity for biodiversity science in countries of origin. Addressing these challenges requires systemic interventions: capacity building in under-represented countries, collaborative frameworks that ensure reciprocity and benefit-sharing, improved digital access, policy reforms to streamline permits without compromising sovereignty and sustained funding for museum science and taxonomy. This study provides a quantitative baseline for monitoring progress towards more equitable specimen curation and offers a reproducible framework applicable to other taxa and regions.