The downstream use cases, benefits, and risks of AI systems depend significantlyon the access afforded to the system, and to whom. However, the downstreamimplications of different access styles are not well understood, making it difficultfor decision-makers to govern model access responsibly. Consequently, we proposeModel Access Governance, a research field focused on helping organisationsand governments make responsible, evidence-based access decisions. We outlinethe motivation for developing this field by highlighting the significant risks ofmisgoverning model access, the limitations of existing research on the topic, andthe opportunity for impact. We then make four sets of recommendations, aimedat helping AI evaluation organisations, frontier AI companies, governments andinternational bodies build consensus around empirically-driven access governance.