From 631aefea18e407922122b7ffc10a6d672801a3a5 Mon Sep 17 00:00:00 2001
From: Harshit28j <Harshitjain0562@gmail.com>
Date: Tue, 10 Mar 2026 16:45:22 +0530
Subject: [PATCH] fix: req changes on feedback from greptile

---
 .../guardrails/guardrail_hooks/presidio.py    |  8 ++++
 .../guardrail_hooks/test_presidio.py          | 45 +++++++++++++++++++
 2 files changed, 53 insertions(+)

diff --git a/litellm/proxy/guardrails/guardrail_hooks/presidio.py b/litellm/proxy/guardrails/guardrail_hooks/presidio.py
index 507a0b89fc9..ddeba2100c3 100644
--- a/litellm/proxy/guardrails/guardrail_hooks/presidio.py
+++ b/litellm/proxy/guardrails/guardrail_hooks/presidio.py
@@ -1150,6 +1150,14 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail):
                         continue
 
                 if not all_chunks:
+                    # All chunks were Anthropic native SSE bytes — output
+                    # masking cannot be applied to raw bytes.  Log a warning
+                    # so operators know PII masking was skipped for this stream.
+                    verbose_proxy_logger.warning(
+                        "Presidio apply_to_output: streaming response contained only "
+                        "bytes chunks (Anthropic native SSE). Output PII masking was "
+                        "skipped for this response."
+                    )
                     return
 
                 assembled_model_response = stream_chunk_builder(
diff --git a/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_presidio.py b/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_presidio.py
index 296faffa1a5..32a8c1b1070 100644
--- a/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_presidio.py
+++ b/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_presidio.py
@@ -2185,3 +2185,48 @@ async def test_apply_guardrail_masks_on_request():
 
     assert "<PERSON>" in result["texts"][0]
     assert "John Smith" not in result["texts"][0]
+
+
+@pytest.mark.asyncio
+async def test_apply_to_output_streaming_bytes_only_logs_warning():
+    """
+    Regression test: when apply_to_output=True and the stream contains only
+    bytes chunks (Anthropic native SSE), output masking is skipped.
+    A warning must be logged so operators are aware.
+    """
+    guardrail = _OPTIONAL_PresidioPIIMasking(
+        mock_testing=True,
+        apply_to_output=True,
+    )
+
+    byte_chunks = [
+        b'data: {"type":"content_block_delta","delta":{"text":"Hello"}}\n\n',
+        b'data: {"type":"content_block_delta","delta":{"text":" world"}}\n\n',
+    ]
+
+    async def mock_stream():
+        for b in byte_chunks:
+            yield b
+
+    mock_user_api_key = UserAPIKeyAuth(api_key="test-key")
+
+    collected = []
+    with patch(
+        "litellm.proxy.guardrails.guardrail_hooks.presidio.verbose_proxy_logger"
+    ) as mock_logger:
+        async for chunk in guardrail.async_post_call_streaming_iterator_hook(
+            user_api_key_dict=mock_user_api_key,
+            response=mock_stream(),
+            request_data={},
+        ):
+            collected.append(chunk)
+
+        # All bytes should be yielded through
+        assert len(collected) == len(byte_chunks)
+        for original, received in zip(byte_chunks, collected):
+            assert original == received
+
+        # Warning must be logged about skipped masking
+        mock_logger.warning.assert_called_once()
+        warning_msg = mock_logger.warning.call_args[0][0]
+        assert "Output PII masking was skipped" in warning_msg