To me this feels like exactly the kind of thing the rule is trying to catch, rather than something we should avoid emitting a diagnostic for: I commented at #11869 (comment). If you want to use the value of the exception in the assertion message, you can do that using pytest.raises():

def allow_exception_in_message():
    x = 0
    with pytest.raises(ZeroDivisionError) as exc_info:
        1 / x
    assert x == 0, exc_info.value

I'm leaning towards agreeing with Alex here