Goodwill doesn't equal transparency or auditability, which are the core concerns that have been repeated around AI models and training of said models