The filter is imperfect — in our tests, an AI asked explicit self-harm questions might type out a fully visible violent answer before the ...
確定！ 回上一頁
