diff --git a/test/transforms/test_env_transforms.py b/test/transforms/test_env_transforms.py
index ce3b61d07ed..2a97ca7e230 100644
--- a/test/transforms/test_env_transforms.py
+++ b/test/transforms/test_env_transforms.py
@@ -501,6 +501,61 @@ def test_stepcounter_marl_truncation(self):
             : max_steps - 1
         ].any(), "Root done should be False before max_steps"
 
+    @pytest.mark.parametrize("truncating_group", ["left", "right"])
+    @pytest.mark.parametrize("update_done", [False, True])
+    def test_stepcounter_independent_nested_groups(self, truncating_group, update_done):
+        class GroupedEnv(EnvBase):
+            def __init__(self):
+                super().__init__()
+                self.observation_spec = Composite(observation=Unbounded(shape=(1,)))
+                self.action_spec = Categorical(2, shape=(1,))
+                self.reward_spec = Unbounded(shape=(1,))
+                flags = Composite(
+                    {
+                        key: Categorical(2, shape=(1,), dtype=torch.bool)
+                        for key in ("done", "terminated", "truncated")
+                    }
+                )
+                group = flags.clone()
+                group["agents"] = flags.expand(3)
+                self.done_spec = Composite(left=group.clone(), right=group.clone())
+
+            def _reset(self, tensordict=None, **kwargs):
+                return self.observation_spec.zero().update(self.full_done_spec.zero())
+
+            def _step(self, tensordict):
+                result = self._reset().update(self.full_reward_spec.zero())
+                result["right", "agents", "truncated"][0] = True
+                result["right", "agents", "done"][0] = True
+                return result
+
+            def _set_seed(self, seed):
+                pass
+
+        env = TransformedEnv(GroupedEnv(), StepCounter(3, update_done=update_done))
+        try:
+            td = env.reset()
+            td[truncating_group, "step_count"].fill_(2)
+            rollout = env.rollout(1, tensordict=td, auto_reset=False)
+            for group in ("left", "right"):
+                truncated = group == truncating_group
+                torch.testing.assert_close(
+                    rollout["next", group, "truncated"], torch.tensor([[truncated]])
+                )
+                agent_done = torch.tensor([group == "right", False, False]).reshape(
+                    1, 3, 1
+                )
+                torch.testing.assert_close(
+                    rollout["next", group, "agents", "truncated"],
+                    agent_done | truncated,
+                )
+                torch.testing.assert_close(
+                    rollout["next", group, "agents", "done"],
+                    agent_done | (truncated and update_done),
+                )
+        finally:
+            env.close()
+
 
 class _AgentNestedCountingEnv(EnvBase):
     """Counting env whose done keys live under the ``agent`` nest."""
diff --git a/torchrl/envs/transforms/_env.py b/torchrl/envs/transforms/_env.py
index e20b1e99077..ff880efc3a4 100644
--- a/torchrl/envs/transforms/_env.py
+++ b/torchrl/envs/transforms/_env.py
@@ -862,6 +862,8 @@ class StepCounter(Transform):
     .. note:: To ensure compatibility with environments that have multiple
         done_key(s), this transform will write a step_count entry for
         every done entry within the tensordict.
+        Truncation is propagated to nested done entries only from their own
+        parent group, not from independent sibling groups.
 
     Examples:
         >>> import gymnasium
@@ -1143,6 +1145,11 @@ def _propagate_to_nested_keys(self, next_tensordict: TensorDictBase) -> None:
                 continue
             # Find a parent truncated key to propagate from
             for parent_key in self.truncated_keys:
+                if (
+                    isinstance(parent_key, tuple)
+                    and nested_key[: len(parent_key) - 1] != parent_key[:-1]
+                ):
+                    continue
                 parent_truncated = next_tensordict.get(parent_key, None)
                 if parent_truncated is not None:
                     # Insert extra dims (e.g. agent dims) so the parent is
@@ -1164,6 +1171,11 @@ def _propagate_to_nested_keys(self, next_tensordict: TensorDictBase) -> None:
                     continue
                 # Find a parent done key to propagate from
                 for parent_key in self.done_keys:
+                    if (
+                        isinstance(parent_key, tuple)
+                        and nested_key[: len(parent_key) - 1] != parent_key[:-1]
+                    ):
+                        continue
                     parent_done = next_tensordict.get(parent_key, None)
                     if parent_done is not None:
                         parent_val = parent_done