check for AWS exceptions despite a 200 response

litellm/proxy/guardrails/guardrail_hooks/bedrock_guardrails.py

@@ -383,6 +383,9 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         )
         #########################################################
         if response.status_code == 200:
+            # check if the response contains an error
+            if self._check_bedrock_response_for_exception(response=response):
+                raise self._get_http_exception_for_failed_guardrail(response)
             # check if the response was flagged
             _json_response = response.json()
             redacted_response = _redact_pii_matches(_json_response)
@@ -403,6 +406,11 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
 
         return bedrock_guardrail_response
 
+    def _check_bedrock_response_for_exception(self, response: httpx.Response) -> bool:
+        return "Exception" in json.loads(response.content.decode("utf-8")).get(
+            "Output", {}
+        ).get("__type", "")
+
     def _get_bedrock_guardrail_response_status(
         self, response: httpx.Response
     ) -> Literal["success", "failure"]:
@@ -410,9 +418,24 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         Get the status of the bedrock guardrail response.
         """
         if response.status_code == 200:
+            if self._check_bedrock_response_for_exception(response):
+                return "failure"
             return "success"
         return "failure"
 
+    def _get_http_exception_for_failed_guardrail(
+        self, response: httpx.Response
+    ) -> HTTPException:
+        return HTTPException(
+            status_code=400,
+            detail={
+                "error": "Guardrail application failed.",
+                "bedrock_guardrail_response": json.loads(
+                    response.content.decode("utf-8")
+                ).get("Output", {}),
+            },
+        )
+
     def _get_http_exception_for_blocked_guardrail(
         self, response: BedrockGuardrailResponse
     ) -> HTTPException:
@@ -562,11 +585,11 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         #########################################################
         ########## 2. Update the messages with the guardrail response ##########
         #########################################################
-        data["messages"] = (
-            self._update_messages_with_updated_bedrock_guardrail_response(
-                messages=new_messages,
-                bedrock_guardrail_response=bedrock_guardrail_response,
-            )
+        data[
+            "messages"
+        ] = self._update_messages_with_updated_bedrock_guardrail_response(
+            messages=new_messages,
+            bedrock_guardrail_response=bedrock_guardrail_response,
         )
 
         #########################################################
@@ -617,11 +640,11 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM):
         #########################################################
         ########## 2. Update the messages with the guardrail response ##########
         #########################################################
-        data["messages"] = (
-            self._update_messages_with_updated_bedrock_guardrail_response(
-                messages=new_messages,
-                bedrock_guardrail_response=bedrock_guardrail_response,
-            )
+        data[
+            "messages"
+        ] = self._update_messages_with_updated_bedrock_guardrail_response(
+            messages=new_messages,
+            bedrock_guardrail_response=bedrock_guardrail_response,
         )
 
         #########################################################

tests/test_litellm/proxy/guardrails/guardrail_hooks/test_bedrock_guardrails.py

@@ -1,12 +1,13 @@
 """
 Unit tests for Bedrock Guardrails
 """
-
+import json
 import os
 import sys
 from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
+from fastapi import HTTPException
 
 sys.path.insert(0, os.path.abspath("../../../../../.."))
 
@@ -238,7 +239,6 @@ async def test_bedrock_guardrail_logging_uses_redacted_response():
     ) as mock_load_creds, patch.object(
         guardrail, "_prepare_request", return_value=MagicMock()
     ) as mock_prepare_request:
-
         mock_post.return_value = mock_bedrock_response
 
         # Call the method that should log the redacted response
@@ -345,7 +345,6 @@ async def test_bedrock_guardrail_original_response_not_modified():
     ) as mock_load_creds, patch.object(
         guardrail, "_prepare_request", return_value=MagicMock()
     ) as mock_prepare_request:
-
         mock_post.return_value = mock_bedrock_response
 
         # Call the method
@@ -860,7 +859,6 @@ async def test__redact_pii_matches_comprehensive_coverage():
 
     print("Comprehensive coverage redaction test passed")
 
-
 @pytest.mark.asyncio
 async def test_bedrock_guardrail_respects_custom_runtime_endpoint(monkeypatch):
     """Test that BedrockGuardrail respects aws_bedrock_runtime_endpoint when set"""
@@ -1049,3 +1047,77 @@ async def test_bedrock_guardrail_parameter_takes_precedence_over_env(monkeypatch
         ), f"Expected parameter endpoint to take precedence. Got: {prepped_request.url}"
 
         print(f"Parameter precedence test passed. URL: {prepped_request.url}")
+
+
+@pytest.mark.asyncio
+async def test_bedrock_guardrail_200_with_exception_in_output_raises_and_logs_failure():
+    """
+    When Bedrock returns HTTP 200 but the body contains Output.__type with 'Exception',
+    the guardrail should:
+      - raise an HTTPException(400) with the Output payload in detail
+      - log the request trace with guardrail_status='failure'
+    """
+    guardrail = BedrockGuardrail(
+        guardrailIdentifier="test-guardrail", guardrailVersion="DRAFT"
+    )
+
+    # Mock a Bedrock "success" HTTP status but an Exception embedded in the body
+    payload = {
+        "Output": {
+            "__type": "com.amazonaws#InternalServerException",
+            "message": "Something went wrong upstream",
+        },
+        "action": "NONE",
+    }
+    mock_resp = MagicMock()
+    mock_resp.status_code = 200
+    mock_resp.content = json.dumps(payload).encode("utf-8")
+    mock_resp.text = json.dumps(payload)
+    mock_resp.json.return_value = payload
+
+    # Minimal request data
+    request_data = {
+        "model": "gpt-4o",
+        "messages": [{"role": "user", "content": "hello"}],
+    }
+
+    # Mock creds and request prep
+    mock_credentials = MagicMock()
+    mock_credentials.access_key = "ak"
+    mock_credentials.secret_key = "sk"
+    mock_credentials.token = None
+
+    with patch.object(
+        guardrail.async_handler, "post", new_callable=AsyncMock
+    ) as mock_post, patch.object(
+        guardrail, "_load_credentials", return_value=(mock_credentials, "us-east-1")
+    ), patch.object(
+        guardrail,
+        "_prepare_request",
+        return_value=MagicMock(url="http://example", headers={}, body=b""),
+    ), patch.object(
+        guardrail, "add_standard_logging_guardrail_information_to_request_data"
+    ) as mock_add_trace:
+        mock_post.return_value = mock_resp
+
+        with pytest.raises(HTTPException) as excinfo:
+            await guardrail.make_bedrock_api_request(
+                source="INPUT",
+                messages=request_data["messages"],
+                request_data=request_data,
+            )
+
+        # 1) Raised HTTPException with 400 status
+        err = excinfo.value
+        assert err.status_code == 400
+        assert err.detail["error"] == "Guardrail application failed."
+
+        # 2) Detail includes the Output object from the Bedrock body
+        assert err.detail["bedrock_guardrail_response"] == payload["Output"]
+
+        # 3) Trace logging received a 'failure' status
+        assert mock_add_trace.called
+        _, kwargs = mock_add_trace.call_args
+        assert kwargs["guardrail_status"] == "failure"
+        # And the JSON passed to tracing is the same response we received
+        assert kwargs["guardrail_json_response"] == payload