Is the concern that prompts could be re-used for training by the provider and such knowledge become part of the model?