diff --git a/python/cuml/cuml/dask/common/base.py b/python/cuml/cuml/dask/common/base.py
index 1f2f71542c..96806a6191 100644
--- a/python/cuml/cuml/dask/common/base.py
+++ b/python/cuml/cuml/dask/common/base.py
@@ -57,7 +57,9 @@ def __init__(self, *, client=None, verbose=False, **kwargs):
         self.internal_model = None
 
     def __getstate__(self):
-        internal_model = self._get_internal_model().result()
+        internal_model = self._get_internal_model()
+        if internal_model:
+            internal_model = internal_model.result()
         state = {
             "verbose": self.verbose,
             "kwargs": self.kwargs,
diff --git a/python/cuml/cuml/tests/dask/test_dask_serialization.py b/python/cuml/cuml/tests/dask/test_dask_serialization.py
index 3dc819b24f..0abc822be5 100644
--- a/python/cuml/cuml/tests/dask/test_dask_serialization.py
+++ b/python/cuml/cuml/tests/dask/test_dask_serialization.py
@@ -80,3 +80,16 @@ def test_serialize_mnmg_model(client):
     unpickled_model = pickle.loads(pickled_model)
 
     assert np.allclose(unpickled_model.coef_, model.coef_)
+
+
+def test_serialize_before_training(client):
+    X, y = make_regression(n_samples=1000, n_features=20, random_state=0)
+    X, y = da.from_array(X), da.from_array(y)
+
+    model = LinearRegression(client=client)
+    pickled_model = pickle.dumps(model)
+    unpickled_model = pickle.loads(pickled_model)
+
+    unpickled_model.client = client
+    unpickled_model.fit(X, y)
+    assert hasattr(unpickled_model, "coef_")