Perspectives: artificial intelligence and scientific autonomy: a policy agenda for emerging research systems

Abstract Artificial intelligence is becoming part of the infrastructure through which scientific knowledge is produced, yet access to computing, data, advanced models and expertise remains highly uneven. This Perspective argues that, for emerging research systems, the central science-policy challenge is not simply AI adoption but scientific autonomy: the capacity to choose, understand, adapt, evaluate and govern the technologies on which research increasingly depends. Scientific autonomy is not technological self-sufficiency; it is strategic interdependence supported by domestic capabilities. Drawing on Brazil as an illustrative case, we propose four dimensions of autonomy—epistemic, infrastructural, human-capability and governance autonomy—and a six-part policy agenda based on shared computing, research-data infrastructure, AI literacy, interdisciplinary networks, adaptive governance and capability-building international cooperation. Without deliberate capacity-building, AI may accelerate discovery while deepening structural asymmetries between research systems.