diff --git a/langfuse/_client/client.py b/langfuse/_client/client.py
index 42d861fd4..1fca04f7c 100644
--- a/langfuse/_client/client.py
+++ b/langfuse/_client/client.py
@@ -4,6 +4,7 @@
 """
 
 import asyncio
+import json
 import logging
 import os
 import re
@@ -1949,6 +1950,20 @@ def create_score(
         environment: Optional[str] = None,
     ) -> None: ...
 
+    def _apply_mask(self, data: Any) -> Any:
+        """Apply the configured mask to data sent outside a span, matching span masking."""
+        if data is None or not self._mask:
+            return data
+        try:
+            return self._mask(data=data)
+        except Exception as e:
+            langfuse_logger.error(
+                "Masking error: Custom mask function threw exception when processing "
+                "data. Using fallback masking. Error: %s",
+                e,
+            )
+            return "<fully masked due to failed mask function>"
+
     def create_score(
         self,
         *,
@@ -2018,6 +2033,9 @@ def create_score(
             return
 
         score_id = score_id or self._create_observation_id()
+        comment = self._apply_mask(comment)
+        if comment is not None and not isinstance(comment, str):
+            comment = json.dumps(comment)
 
         try:
             new_body = ScoreBody(
@@ -3550,7 +3568,7 @@ def create_dataset(
             result = self.api.datasets.create(
                 name=name,
                 description=description,
-                metadata=metadata,
+                metadata=self._apply_mask(metadata),
                 input_schema=input_schema,
                 expected_output_schema=expected_output_schema,
             )
@@ -3649,9 +3667,9 @@ def create_dataset_item(
 
             result = self.api.dataset_items.create(
                 dataset_name=dataset_name,
-                input=input,
-                expected_output=expected_output,
-                metadata=metadata,
+                input=self._apply_mask(input),
+                expected_output=self._apply_mask(expected_output),
+                metadata=self._apply_mask(metadata),
                 source_trace_id=source_trace_id,
                 source_observation_id=source_observation_id,
                 status=status,
diff --git a/tests/unit/test_mask_api_payloads.py b/tests/unit/test_mask_api_payloads.py
new file mode 100644
index 000000000..a0237c656
--- /dev/null
+++ b/tests/unit/test_mask_api_payloads.py
@@ -0,0 +1,53 @@
+from unittest.mock import Mock
+
+from langfuse import Langfuse
+
+
+def _client():
+    client = Langfuse(
+        public_key="pk",
+        secret_key="sk",
+        host="https://mock-host.com",
+        tracing_enabled=False,
+        mask=lambda data: "masked",
+    )
+    client.api = Mock()
+    client._resources.add_score_task = Mock()
+    return client
+
+
+def test_create_dataset_item_masks_payload():
+    client = _client()
+    client.create_dataset_item(
+        dataset_name="ds", input="secret", expected_output="secret", metadata="secret"
+    )
+    kwargs = client.api.dataset_items.create.call_args.kwargs
+    assert (kwargs["input"], kwargs["expected_output"], kwargs["metadata"]) == (
+        "masked",
+        "masked",
+        "masked",
+    )
+
+
+def test_create_dataset_masks_metadata():
+    client = _client()
+    client.create_dataset(name="ds", metadata="secret")
+    assert client.api.datasets.create.call_args.kwargs["metadata"] == "masked"
+
+
+def test_create_score_masks_comment():
+    client = _client()
+    client._tracing_enabled = True  # create_score is a no-op otherwise
+    client.create_score(name="s", value=1, trace_id="t" * 32, comment="secret")
+    assert (
+        client._resources.add_score_task.call_args.args[0]["body"].comment == "masked"
+    )
+
+
+def test_create_score_serializes_non_string_masked_comment():
+    client = _client()
+    client._tracing_enabled = True
+    client._mask = lambda data: {"redacted": True}
+    client.create_score(name="s", value=1, trace_id="t" * 32, comment="secret")
+    body = client._resources.add_score_task.call_args.args[0]["body"]
+    assert body.comment == '{"redacted": true}'