Responsible Governance of Generative AI Use Across the Scientific Research Lifecycle: An Integrative Review and Thematic Synthesis of Evidence

Generative artificial intelligence (GenAI) is increasingly embedded across the research lifecycle, raising governance questions about how it shapes epistemic, data-related, and organisational decisions. This study develops an actionable framework for responsible GenAI governance through a structured integrative review updated through 25 August 2026. After source-identity and claim-level verification, 67 publications and 10 policy or integrity documents were included. Evidence was extracted by research stage, risk, governance implication, and responsible actor. Thematic synthesis identified four themes: epistemic and methodological validity; originality, authorship and accountability; data, confidentiality, security and fairness; and institutional governance and capacity. A cross-cutting theme concerned proportionality to research stage, material effect, verifiability, and delegation of human judgement. Multi-coding showed that 60 sources addressed authorship/accountability, 44 governance/institutional capacity, 38 epistemic validity, and 29 data/fairness. Generic disclosure or blanket prohibition was insufficient. Governance intensity should increase with data sensitivity, effects on scientific inference, verification difficulty, and delegation of judgement. A five-dimensional framework links research stage, risk domain, risk severity, required control, and accountable actor. Mature governance requires secure infrastructure, traceability, independent verification, AI literacy, policy review, and human accountability.