diff --git a/bindings/pydrake/gym/_drake_gym_env.py b/bindings/pydrake/gym/_drake_gym_env.py
index b6187615b66e..8a335c94a426 100644
--- a/bindings/pydrake/gym/_drake_gym_env.py
+++ b/bindings/pydrake/gym/_drake_gym_env.py
@@ -33,8 +33,8 @@ def __init__(
         self,
         simulator: Simulator | Callable[[RandomGenerator], Simulator],
         time_step: float,
-        action_space: gym.spaces.Space,
-        observation_space: gym.spaces.Space,
+        action_space: gym.spaces.Space | None,
+        observation_space: gym.spaces.Space | None,
         reward: Callable[[System, Context], float] | OutputPortIndex | str,
         action_port_id: InputPort | InputPortIndex | str = None,
         observation_port_id: OutputPortIndex | str = None,
@@ -127,10 +127,12 @@ def __init__(
         assert time_step > 0
         self.time_step = time_step
 
-        assert isinstance(action_space, gym.spaces.Space)
+        if action_space is not None:
+            assert isinstance(action_space, gym.spaces.Space)
         self.action_space = action_space
 
-        assert isinstance(observation_space, gym.spaces.Space)
+        if observation_space is not None:
+            assert isinstance(observation_space, gym.spaces.Space)
         self.observation_space = observation_space
 
         if isinstance(reward, (OutputPortIndex, str)):
@@ -193,9 +195,21 @@ def _setup(self):
             else:
                 self.action_port = system.GetInputPort(self.action_port_id)
         if self.action_port.get_data_type() == PortDataType.kVectorValued:
+            if self.action_space is None:
+                self.action_space = gym.spaces.Box(
+                    low=-np.inf,
+                    high=np.inf,
+                    shape=(self.action_port.size(),),
+                    dtype=np.float64,
+                )
             assert np.array_equal(
                 self.action_space.shape, [self.action_port.size()]
             )
+        else:
+            assert self.action_space is not None, (
+                "action_space must be provided when the action port is not "
+                "vector-valued"
+            )
 
         def get_output_port(id):
             if isinstance(id, OutputPortIndex):
@@ -206,9 +220,21 @@ def get_output_port(id):
         if self.observation_port_id:
             self.observation_port = get_output_port(self.observation_port_id)
         if self.observation_port.get_data_type() == PortDataType.kVectorValued:
+            if self.observation_space is None:
+                self.observation_space = gym.spaces.Box(
+                    low=-np.inf,
+                    high=np.inf,
+                    shape=(self.observation_port.size(),),
+                    dtype=np.float64,
+                )
             assert np.array_equal(
                 self.observation_space.shape, [self.observation_port.size()]
             )
+        else:
+            assert self.observation_space is not None, (
+                "observation_space must be provided when the observation "
+                "port is not vector-valued"
+            )
 
         # Note: We require that there is no direct feedthrough action_port to
         # observation_port.  Unfortunately, HasDirectFeedthrough returns false
diff --git a/bindings/pydrake/gym/test/drake_gym_test.py b/bindings/pydrake/gym/test/drake_gym_test.py
index 507de50e1049..3aaf4de178df 100644
--- a/bindings/pydrake/gym/test/drake_gym_test.py
+++ b/bindings/pydrake/gym/test/drake_gym_test.py
@@ -1,8 +1,15 @@
 import unittest
 
 import gymnasium as gym
+import numpy as np
 import stable_baselines3.common.env_checker
 
+from pydrake.common.value import Value
+from pydrake.gym import DrakeGymEnv
+from pydrake.systems.analysis import Simulator
+from pydrake.systems.framework import DiagramBuilder
+from pydrake.systems.primitives import PassThrough
+
 
 class DrakeGymTest(unittest.TestCase):
     """
@@ -65,3 +72,70 @@ def test_step(self):
         dut.reset()
         observation, _, _, _, _ = dut.step(dut.action_space.sample())
         self.assertTrue(dut.observation_space.contains(observation))
+
+    def test_none_spaces_default_to_infinite_box(self):
+        """Passing None for vector-valued ports builds ±inf Boxes."""
+        size = 3
+        builder = DiagramBuilder()
+        plant = builder.AddSystem(PassThrough(vector_size=size))
+        builder.ExportInput(plant.get_input_port(), "actions")
+        builder.ExportOutput(plant.get_output_port(), "observations")
+        diagram = builder.Build()
+        simulator = Simulator(diagram)
+
+        dut = DrakeGymEnv(
+            simulator=simulator,
+            time_step=0.1,
+            action_space=None,
+            observation_space=None,
+            reward=lambda system, context: 0.0,
+            action_port_id="actions",
+            observation_port_id="observations",
+        )
+
+        self.assertIsInstance(dut.action_space, gym.spaces.Box)
+        self.assertEqual(dut.action_space.shape, (size,))
+        self.assertTrue(np.all(np.isneginf(dut.action_space.low)))
+        self.assertTrue(np.all(np.isposinf(dut.action_space.high)))
+
+        self.assertIsInstance(dut.observation_space, gym.spaces.Box)
+        self.assertEqual(dut.observation_space.shape, (size,))
+        self.assertTrue(np.all(np.isneginf(dut.observation_space.low)))
+        self.assertTrue(np.all(np.isposinf(dut.observation_space.high)))
+
+    def test_none_spaces_reject_non_vector_ports(self):
+        """None is only valid for vector-valued ports."""
+        builder = DiagramBuilder()
+        system = builder.AddSystem(
+            PassThrough(abstract_model_value=Value("model"))
+        )
+        builder.ExportInput(system.get_input_port(), "actions")
+        builder.ExportOutput(system.get_output_port(), "observations")
+        diagram = builder.Build()
+        simulator = Simulator(diagram)
+        # Any concrete Space is fine here; we only exercise the None path.
+        dummy_space = gym.spaces.Discrete(1)
+
+        with self.assertRaisesRegex(AssertionError, "action_space must be"):
+            DrakeGymEnv(
+                simulator=simulator,
+                time_step=0.1,
+                action_space=None,
+                observation_space=dummy_space,
+                reward=lambda system, context: 0.0,
+                action_port_id="actions",
+                observation_port_id="observations",
+            )
+
+        with self.assertRaisesRegex(
+            AssertionError, "observation_space must be"
+        ):
+            DrakeGymEnv(
+                simulator=simulator,
+                time_step=0.1,
+                action_space=dummy_space,
+                observation_space=None,
+                reward=lambda system, context: 0.0,
+                action_port_id="actions",
+                observation_port_id="observations",
+            )