I don't understand how OpenAI claims it would have happened. The weights are closed and as far as I read they are not complaining Deepseek hacked them and obtained the weight. So all they could do was to query OpenAI and generate test data. But how much did they query really - I would suppose it would require a huge amount done via an external, paid-for API? Is there any proof of this besides OpenAI saying it? Even if we suppose it is true, I suppose this must have happened via the API so they paid per token etc. So they paid for each and every token of training data. As I understand, the requester owns the copyright on what is generated by OpenAI's models and is free to do what they want.