diff --git a/litellm/proxy/utils.py b/litellm/proxy/utils.py
index c095586b6c94..5bfd7f5d05fe 100644
--- a/litellm/proxy/utils.py
+++ b/litellm/proxy/utils.py
@@ -429,6 +429,14 @@ def _enrich_http_exception_with_guardrail_context(exc: BaseException, callback:
         detail.setdefault("guardrail_mode", event_hook)
 
 
+def _is_client_error_exception(exc: Exception) -> bool:
+    if isinstance(exc, HTTPException):
+        return exc.status_code < 500
+    if isinstance(exc, ProxyException):
+        return not (exc.code.isdigit() and int(exc.code) >= 500)
+    return False
+
+
 def _exception_changes_request_flow(exc: BaseException) -> bool:
     """
     True for guardrail exceptions the proxy turns into an alternate request flow
@@ -2885,9 +2893,7 @@ async def post_call_failure_hook(
 
         ### ALERTING ###
         await self.update_request_status(litellm_call_id=request_data.get("litellm_call_id", ""), status="fail")
-        if AlertType.llm_exceptions in self.alert_types and not isinstance(
-            original_exception, (HTTPException, ProxyException)
-        ):
+        if AlertType.llm_exceptions in self.alert_types and not _is_client_error_exception(original_exception):
             """
             Just alert on LLM API exceptions. Do not alert on user errors
 
diff --git a/tests/test_litellm/proxy/test_proxy_utils.py b/tests/test_litellm/proxy/test_proxy_utils.py
index 9def21c0573a..94ccc2762c5c 100644
--- a/tests/test_litellm/proxy/test_proxy_utils.py
+++ b/tests/test_litellm/proxy/test_proxy_utils.py
@@ -13,7 +13,7 @@
 from litellm.types.guardrails import GuardrailEventHooks
 
 
-from unittest.mock import MagicMock, patch
+from unittest.mock import AsyncMock, MagicMock, patch
 
 from litellm.proxy.utils import get_custom_url, join_paths
 
@@ -1303,12 +1303,10 @@ class TestPostCallFailureHookLLMExceptionAlerting:
     """The llm_exceptions alert is for infra / LLM-API failures, not user
     errors (https://github.com/BerriAI/litellm/issues/3395). Already-normalized
     client errors must be excluded so a guardrail content-policy block never
-    pages on-call. ProxyException is such an error; before LIT-3751 only
-    HTTPException was excluded, so AIM blocks paged as if the LLM API failed."""
+    pages on-call. 5xx proxy errors still alert."""
 
-    async def _alerted(self, exc) -> bool:
+    async def _alerted(self, exc: Exception) -> AsyncMock:
         import asyncio
-        from unittest.mock import AsyncMock
 
         from litellm.proxy._types import AlertType, UserAPIKeyAuth
 
@@ -1325,7 +1323,7 @@ async def _alerted(self, exc) -> bool:
                 user_api_key_dict=UserAPIKeyAuth(),
             )
         await asyncio.sleep(0)  # let the fire-and-forget alert task run
-        return alerting_handler.called
+        return alerting_handler
 
     @pytest.mark.asyncio
     async def test_proxy_exception_does_not_alert(self):
@@ -1338,15 +1336,49 @@ async def test_proxy_exception_does_not_alert(self):
             code=400,
             openai_code="content_policy_violation",
         )
-        assert await self._alerted(exc) is False
+        assert (await self._alerted(exc)).called is False
 
     @pytest.mark.asyncio
     async def test_http_exception_does_not_alert(self):
-        assert await self._alerted(HTTPException(status_code=400, detail="blocked")) is False
+        assert (await self._alerted(HTTPException(status_code=400, detail="blocked"))).called is False
 
     @pytest.mark.asyncio
     async def test_genuine_llm_api_error_still_alerts(self):
-        assert await self._alerted(Exception("upstream 503")) is True
+        assert (await self._alerted(Exception("upstream 503"))).called is True
+
+    @pytest.mark.asyncio
+    async def test_http_exception_5xx_alerts(self):
+        alerting_handler = await self._alerted(
+            HTTPException(
+                status_code=502,
+                detail={
+                    "error": "Headroom compression service returned an error",
+                    "status_code": 503,
+                    "guardrail_name": "headroom-compression-global",
+                },
+            )
+        )
+        assert alerting_handler.called is True
+        assert "headroom-compression-global" in alerting_handler.call_args.kwargs["message"]
+
+    @pytest.mark.asyncio
+    async def test_proxy_exception_5xx_alerts(self):
+        from litellm.proxy._types import ProxyException
+
+        alerting_handler = await self._alerted(
+            ProxyException(
+                message="guardrail backend down",
+                type="internal_server_error",
+                param=None,
+                code=503,
+            )
+        )
+        assert alerting_handler.called is True
+
+    @pytest.mark.asyncio
+    async def test_http_exception_429_does_not_alert(self):
+        alerting_handler = await self._alerted(HTTPException(status_code=429, detail="rate limited"))
+        assert alerting_handler.called is False
 
 
 class TestPostCallFailureHookProxyExceptionLogging: