From ffe5d303e5294cdcfb0db43d32e3ca385f141a41 Mon Sep 17 00:00:00 2001
From: Yucheng He <yucheng@berri.ai>
Date: Fri, 18 Sep 2026 01:30:46 -0700
Subject: [PATCH] fix(llmguard): accept proxy async call types

---
 .../enterprise_callbacks/llm_guard.py         | 18 +++-
 tests/local_testing/test_llm_guard.py         | 93 +++++++++++++++++++
 2 files changed, 107 insertions(+), 4 deletions(-)

diff --git a/enterprise/litellm_enterprise/enterprise_callbacks/llm_guard.py b/enterprise/litellm_enterprise/enterprise_callbacks/llm_guard.py
index d10b5a2ab09..9c8537e6820 100644
--- a/enterprise/litellm_enterprise/enterprise_callbacks/llm_guard.py
+++ b/enterprise/litellm_enterprise/enterprise_callbacks/llm_guard.py
@@ -8,7 +8,7 @@
 ## This provides an LLM Guard Integration for content moderation on the proxy
 
 import asyncio
-from typing import Optional
+from typing import Final, Optional
 
 import aiohttp
 from fastapi import HTTPException
@@ -137,15 +137,25 @@ class _ENTERPRISE_LLMGuard(CustomLogger):
             return
 
         self.print_verbose("Makes LLM Guard Check")
-        if call_type not in [
+        accepted_call_types: Final = (
             "completion",
+            "acompletion",
+            "text_completion",
+            "atext_completion",
             "embeddings",
+            "embedding",
+            "aembedding",
             "image_generation",
+            "aimage_generation",
             "moderation",
+            "amoderation",
             "audio_transcription",
-        ]:
+            "transcription",
+            "atranscription",
+        )
+        if call_type not in accepted_call_types:
             self.print_verbose(
-                f"Call Type - {call_type}, not in accepted list - ['completion','embeddings','image_generation','moderation','audio_transcription']"
+                f"Call Type - {call_type}, not in accepted list - {accepted_call_types}"
             )
             return data
 
diff --git a/tests/local_testing/test_llm_guard.py b/tests/local_testing/test_llm_guard.py
index 9e70d48dbda..ceb77386349 100644
--- a/tests/local_testing/test_llm_guard.py
+++ b/tests/local_testing/test_llm_guard.py
@@ -5,6 +5,7 @@
 ## Unit test for presidio pii masking
 import sys, os, asyncio, time, random
 from datetime import datetime
+from typing import Final, Literal
 import traceback
 from dotenv import load_dotenv
 
@@ -19,6 +20,7 @@ from litellm import Router, mock_completion
 from litellm.proxy.utils import ProxyLogging, hash_token
 from litellm.proxy._types import UserAPIKeyAuth
 from litellm.caching.caching import DualCache
+from litellm.types.utils import CallTypesLiteral
 
 ### UNIT TESTS FOR LLM GUARD ###
 
@@ -106,6 +108,97 @@ async def test_llm_guard_sanitizes_multimodal_and_input():
     assert input_result["input"] == ["email: [REDACTED]", "email: [REDACTED]"]
 
 
+@pytest.mark.parametrize(
+    "call_type, payload_key",
+    (
+        ("completion", "messages"),
+        ("acompletion", "messages"),
+        ("text_completion", "prompt"),
+        ("atext_completion", "prompt"),
+        ("embeddings", "input"),
+        ("embedding", "input"),
+        ("aembedding", "input"),
+        ("moderation", "input"),
+        ("amoderation", "input"),
+        ("image_generation", "prompt"),
+        ("aimage_generation", "prompt"),
+        ("audio_transcription", "prompt"),
+        ("transcription", "prompt"),
+        ("atranscription", "prompt"),
+    ),
+)
+@pytest.mark.parametrize("is_valid", (True, False))
+@pytest.mark.asyncio
+async def test_llm_guard_call_type_aliases(
+    call_type: CallTypesLiteral,
+    payload_key: Literal["messages", "input", "prompt"],
+    is_valid: bool,
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(litellm, "llm_guard_mode", "all")
+    llm_guard: Final = _ENTERPRISE_LLMGuard(
+        mock_testing=True,
+        mock_redacted_text={
+            "sanitized_prompt": "email: [REDACTED]",
+            "is_valid": is_valid,
+        },
+    )
+    user_api_key_dict: Final = UserAPIKeyAuth(api_key=hash_token("sk-12345"))
+    data: Final = {
+        payload_key: [{"role": "user", "content": "email: person@example.com"}]
+        if payload_key == "messages"
+        else "email: person@example.com"
+    }
+
+    if not is_valid:
+        with pytest.raises(HTTPException) as exc_info:
+            await llm_guard.async_moderation_hook(
+                data=data, user_api_key_dict=user_api_key_dict, call_type=call_type
+            )
+        assert exc_info.value.status_code == 400
+        assert exc_info.value.detail == {"error": "Violated content safety policy"}
+        return
+
+    result: Final = await llm_guard.async_moderation_hook(
+        data=data, user_api_key_dict=user_api_key_dict, call_type=call_type
+    )
+    assert result is data
+    assert data[payload_key] == (
+        [{"role": "user", "content": "email: [REDACTED]"}]
+        if payload_key == "messages"
+        else "email: [REDACTED]"
+    )
+
+
+@pytest.mark.parametrize(
+    "call_type",
+    (
+        "responses",
+        "aresponses",
+        "anthropic_messages",
+        "aanthropic_messages",
+        "aspeech",
+        "aimage_edit",
+        "pass_through_endpoint",
+    ),
+)
+@pytest.mark.asyncio
+async def test_llm_guard_skips_unsupported_call_types(
+    call_type: CallTypesLiteral, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    monkeypatch.setattr(litellm, "llm_guard_mode", "all")
+    llm_guard: Final = _ENTERPRISE_LLMGuard(
+        mock_testing=True,
+        mock_redacted_text={"is_valid": False},
+    )
+    data: Final = {"messages": [{"role": "user", "content": "unchanged"}]}
+    result: Final = await llm_guard.async_moderation_hook(
+        data=data, user_api_key_dict=UserAPIKeyAuth(), call_type=call_type
+    )
+    assert result is data
+    assert data == {"messages": [{"role": "user", "content": "unchanged"}]}
+
+
 @pytest.mark.asyncio
 async def test_llm_guard_error_raising():
     """