diff --git a/litellm/proxy/utils.py b/litellm/proxy/utils.py
index d3ca2fa64ed..be29b4ec5b6 100644
--- a/litellm/proxy/utils.py
+++ b/litellm/proxy/utils.py
@@ -88,7 +88,10 @@ from litellm.integrations.custom_logger import CustomLogger
 from litellm.integrations.prometheus import PrometheusLogger
 from litellm.integrations.SlackAlerting.slack_alerting import SlackAlerting
 from litellm.integrations.SlackAlerting.utils import _add_langfuse_trace_id_to_alert
-from litellm.litellm_core_utils.core_helpers import coerce_token_limit
+from litellm.litellm_core_utils.core_helpers import (
+    coerce_token_limit,
+    get_metadata_variable_name_from_kwargs,
+)
 from litellm.litellm_core_utils.litellm_logging import Logging
 from litellm.litellm_core_utils.safe_json_dumps import safe_dumps
 from litellm.litellm_core_utils.safe_json_loads import safe_json_loads
@@ -1248,7 +1251,7 @@ class ProxyLogging:
 
         Returns the (possibly modified) data dict.
         """
-        metadata: Final = data.get("metadata", data.get("litellm_metadata", {})) or {}
+        metadata: Final = data.get(get_metadata_variable_name_from_kwargs(data)) or {}
         pipelines: Final = metadata.get("_guardrail_pipelines")
         if not pipelines:
             return data
@@ -1400,7 +1403,7 @@ class ProxyLogging:
             )
 
             # Get pipeline-managed guardrails to skip in normal loop
-            metadata: Final = data.get("metadata", data.get("litellm_metadata", {})) or {}
+            metadata: Final = data.get(get_metadata_variable_name_from_kwargs(data)) or {}
             pipeline_managed: Final[set] = metadata.get("_pipeline_managed_guardrails", set())
 
             caps: Final = ProxyLogging._callback_capabilities()
diff --git a/tests/test_litellm/proxy/utils/proxy_logging/test_guardrail_pipeline.py b/tests/test_litellm/proxy/utils/proxy_logging/test_guardrail_pipeline.py
index 5c711fc6c34..98eeec21dab 100644
--- a/tests/test_litellm/proxy/utils/proxy_logging/test_guardrail_pipeline.py
+++ b/tests/test_litellm/proxy/utils/proxy_logging/test_guardrail_pipeline.py
@@ -350,6 +350,44 @@ async def test_maybe_execute_pipelines_skips_pipelines_with_other_mode(proxy_log
     assert out is data
 
 
+@pytest.mark.asyncio
+async def test_maybe_execute_pipelines_reads_litellm_metadata_when_caller_sends_own_metadata(
+    proxy_logging, make_user_api_key_auth, monkeypatch
+):
+    """On /v1/messages the proxy stores policy state in ``litellm_metadata``, while the
+    caller's provider-facing ``metadata`` (Claude Code sends ``metadata.user_id``) stays
+    untouched. The pipeline must still run."""
+    pipeline = MagicMock()
+    pipeline.mode = "pre_call"
+    pipeline.steps = []
+    fake_result = MagicMock()
+    fake_result.terminal_action = "allow"
+    fake_result.modified_data = None
+    fake_result.step_results = []
+    data = {
+        "metadata": {"user_id": "user_abc"},
+        "litellm_metadata": {"_guardrail_pipelines": [("policy-1", pipeline)]},
+        "messages": [],
+        "model": "m",
+    }
+    executed = MagicMock(return_value=fake_result)
+
+    async def fake_execute_steps(**kwargs):
+        return executed(**kwargs)
+
+    monkeypatch.setattr(
+        "litellm.proxy.policy_engine.pipeline_executor.PipelineExecutor.execute_steps",
+        fake_execute_steps,
+    )
+    await proxy_logging._maybe_execute_pipelines(
+        data=data,
+        user_api_key_dict=make_user_api_key_auth(),
+        call_type="anthropic_messages",
+        event_hook="pre_call",
+    )
+    executed.assert_called_once()
+
+
 @pytest.mark.asyncio
 async def test_maybe_execute_pipelines_blocks_on_block_terminal_action_raises(
     proxy_logging, make_user_api_key_auth, monkeypatch