The Trump administration has finalized a voluntary framework for testing new artificial intelligence models for cybersecurity and safety risks, but plans to keep the policy details private rather than releasing them publicly.