diff --git a/keras/src/layers/activations/softmax.py b/keras/src/layers/activations/softmax.py
index c1fee581a89..90884e50a86 100644
--- a/keras/src/layers/activations/softmax.py
+++ b/keras/src/layers/activations/softmax.py
@@ -46,7 +46,6 @@ def __init__(self, axis=-1, **kwargs):
         super().__init__(**kwargs)
         self.supports_masking = True
         self.axis = axis
-        self.built = True
 
     def call(self, inputs, mask=None):
         if mask is not None:
diff --git a/keras/src/testing/test_case.py b/keras/src/testing/test_case.py
index 9651eab71c0..351f58bcb91 100644
--- a/keras/src/testing/test_case.py
+++ b/keras/src/testing/test_case.py
@@ -271,6 +271,33 @@ def run_layer_test(
                 lambda _: "float32", input_shape
             )
 
+        # Estimate actual number of weights, variables, seed generators if
+        # expected ones not set. When using layers uses composition it should
+        # build each sublayer manually.
+        if input_data is not None or input_shape is not None:
+            if input_data is None:
+                input_data = create_eager_tensors(
+                    input_shape, input_dtype, input_sparse
+                )
+            layer = layer_cls(**init_kwargs)
+            if isinstance(input_data, dict):
+                layer(**input_data, **call_kwargs)
+            else:
+                layer(input_data, **call_kwargs)
+
+            if expected_num_trainable_weights is None:
+                expected_num_trainable_weights = len(layer.trainable_weights)
+            if expected_num_non_trainable_weights is None:
+                expected_num_non_trainable_weights = len(
+                    layer.non_trainable_weights
+                )
+            if expected_num_non_trainable_variables is None:
+                expected_num_non_trainable_variables = len(
+                    layer.non_trainable_variables
+                )
+            if expected_num_seed_generators is None:
+                expected_num_seed_generators = len(get_seed_generators(layer))
+
         # Serialization test.
         layer = layer_cls(**init_kwargs)
         self.run_class_serialization_test(layer, custom_objects)