It may be worth editing your post to specify that you’re talking about LLMs. There’s no indication that this is your intent without reading your responses in the comment section.

Ethical meaning : “private”, "anonymous, “not training with your data”, “no censured”, “open source”…

Private, anonymous, and uncensored. Those are easy. There’s plenty of pre-trained LLMs out there that you can download and use however you like.

Not training with your data, not possible as far as I’m aware. LLMs rely on the availability of a huge quantity and diversity of data. There isn’t enough of that available that also come with consent of the creator for this usage.

Open source isn’t well defined for machine learning models. Lots of models have their code and weights available, so if that qualifies for you, that’s also easy to find. Huggingface hosts most of them.