Evidence synthesis and decision governance for artificial intelligence in sustainable gold mining in the Ecuadorian Amazon

Abstract Artificial intelligence is increasingly used across mining, but its contribution to sustainable gold production remains uncertain. This integrative review evaluates where AI evidence is mature, how it is validated, and what it implies for Ecuador and the Chupianza concession. The revised auditable corpus comprised 42 scientific and technical sources published from 2014 to September 2026 and coded by lifecycle stage, region, method, validation, sustainability dimension and evidence level. Monitoring and environmental-context studies were most frequent ( n = 17), followed by exploration and resources ( n = 11), processing and recovery ( n = 8), and planning, operations and safety ( n = 6). Eighteen sources evaluated specific AI models, but none prospectively demonstrated improvement in operational decisions or measured sustainability outcomes. Ecuadorian evidence mainly documents mercury exposure, land degradation, restoration needs and remotely observed mining-associated change. A four-layer evidence-to-decision framework is proposed to connect local data, models, accountable decisions and economic, environmental, safety and social outcomes. Chupianza is a prospective test setting rather than a demonstrated AI application; geological characterization, baseline indicators, spatial and temporal validation, uncertainty controls and human oversight are required before deployment claims can be made.