Spot the odd one out:

• np.array defaults to copy=True, subok=False
• np.ma.array defaults to copy=False, subok=True

• np.asarray defaults to copy=False, subok=False
• np.ma.asarray defaults to copy=False, subok=False

• np.asanyarray defaults to copy=False, subok=True
• np.ma.asanyarray defaults to copy=False, subok=True

So as a result, np.ma.asanyarray and np.ma.array are exactly the same.
More confusingly, np.ma.asarray is stricter than np.ma.array, not less strict!

I don't know if there's any way we can fix this without breaking downstream code, but it seems like a poor design decision - and violates the expectation that np.ma.func and np.func are similar.