What does the term "black$-$box attack" refer to in adversarial machine learning?

Question $39$ options:

Attacks with no knowledge of the model's parameters

Attacks with full access to the model's parameters

Attacks that manipulate training data

Attacks that infer training data