I’m fairly sure that the actual lesson of the recent Anthropic brouhaha is that AI models cannot be made safe. Even a thorough testing and licensing regime would leave ample space for rogue actors to leverage open source models. That genie is thoroughly out of the bottle.