I think the genie is out of the bottle—I'm not sure how you could prevent consumers from training open models using open source code, freely available datasets and GPUs that are already deployed…