A groundswell of interest in how America’s top AI companies train their models and how they try to prevent cybersecurity incidents has focused attention on the lack of AI regulation.