There are different levels of trust one can apply. In the case of Deepseek you can run the model on your very own hardware with no external connection to China at all. You can abliterate and fine-tune the model's weights yourself. You can observe all of its reasoning text as it works.

So I think I'd trust Deepseek over the likes of OpenAI, who insist you send all your data to them and be happy with whatever they choose to send back to you.