diff --git a/python/mxnet/ndarray/ndarray.py b/python/mxnet/ndarray/ndarray.py
index 1c182731c78e..232589063ff0 100644
--- a/python/mxnet/ndarray/ndarray.py
+++ b/python/mxnet/ndarray/ndarray.py
@@ -4212,7 +4212,12 @@ def dl_managed_tensor_deleter(dl_managed_tensor_handle):
 
 
 def from_numpy(ndarray, zero_copy=True):
-    """Returns an MXNet's NDArray backed by Numpy's ndarray.
+    """Returns an MXNet's ndarray backed by numpy's ndarray.
+    When `zero_copy` is set to be true,
+    this API consumes numpy's ndarray and produces MXNet's ndarray
+    without having to copy the content. In this case, we disallow
+    users to modify the given numpy ndarray, and it is suggested
+    not to read the numpy ndarray as well for internal correctness.
 
     Parameters
     ----------
@@ -4261,6 +4266,7 @@ def _make_dl_managed_tensor(array):
 
     if not ndarray.flags['C_CONTIGUOUS']:
         raise ValueError("Only c-contiguous arrays are supported for zero-copy")
+    ndarray.flags['WRITEABLE'] = False
     c_obj = _make_dl_managed_tensor(ndarray)
     address = ctypes.addressof(c_obj)
     address = ctypes.cast(address, ctypes.c_void_p)
diff --git a/tests/python/unittest/test_ndarray.py b/tests/python/unittest/test_ndarray.py
index c62bd19453d9..df505436fa0c 100644
--- a/tests/python/unittest/test_ndarray.py
+++ b/tests/python/unittest/test_ndarray.py
@@ -1687,8 +1687,8 @@ def test_zero_from_numpy():
             mx.test_utils.assert_almost_equal(np_array, mx_array.asnumpy())
     np_array = arrays[0]
     mx_array = mx.nd.from_numpy(np_array)
-    np_array[2, 1] = 0
-    mx.test_utils.assert_almost_equal(np_array, mx_array.asnumpy())
+    assertRaises(ValueError, np_array.__setitem__, (2, 1), 0)
+
     mx_array[2, 1] = 100
     mx.test_utils.assert_almost_equal(np_array, mx_array.asnumpy())
     np_array = np.array([[1, 2], [3, 4], [5, 6]]).transpose()