Abstract
AI developers and policy makers need to devise methods of commercial model deprecation that balance developers’ reasonable interest in restricting access to outdated models with the value of preserving technological developments and enabling ongoing scientific research across versions.