Yes, it is possible. The type of organization developing the project may vary. Companies such as DeepSeek, Meta, and similar ones provide the model architecture and weights, ready for use.

That said, the process is not fully transparent: as an investor in this context, you do not know which data the model was trained on, nor for what purpose. Were training prompts used to enforce certain behaviors? To bias the model against specific ideals?

This should be auditable in my opinion. Then, If fine tuning with private data is required afterward, each user providing this data should have their own secure copy of the model.