From 1b8b17cc8a911e31be0522d8ab57e647e80069bc Mon Sep 17 00:00:00 2001
From: yucheng <yucheng@berri.ai>
Date: Mon, 14 Sep 2026 23:54:15 +0000
Subject: [PATCH] fix(guardrails): judge only the latest request turn in
 pre_call and during_call llm_as_a_judge

The request-side prompt told the judge to focus on the most recent user turn but the text under review was every extracted request message joined together, so a multi-turn request with an off-topic earlier turn and an on-topic latest turn scored 50 and was blocked. Request-side judging now evaluates the last extracted request text (after the configured message scoping) and passes the full role-labelled conversation only as context. Response-side judging still evaluates all extracted response text

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 .../llm_as_a_judge/__init__.py                | 11 +++++++---
 .../proxy/guardrails/test_llm_as_a_judge.py   | 22 +++++++++++++++----
 2 files changed, 26 insertions(+), 7 deletions(-)

diff --git a/litellm/proxy/guardrails/guardrail_hooks/llm_as_a_judge/__init__.py b/litellm/proxy/guardrails/guardrail_hooks/llm_as_a_judge/__init__.py
index d00bf86ed2b..afb39d53393 100644
--- a/litellm/proxy/guardrails/guardrail_hooks/llm_as_a_judge/__init__.py
+++ b/litellm/proxy/guardrails/guardrail_hooks/llm_as_a_judge/__init__.py
@@ -54,7 +54,7 @@ JUDGE_SYSTEM_PROMPTS: Final[MappingProxyType[JudgeInputType, str]] = MappingProx
 )
 
 _JUDGE_SUBJECT_LABELS: Final[MappingProxyType[JudgeInputType, str]] = MappingProxyType(
-    {"request": "Request text to evaluate", "response": "Assistant response to evaluate"}
+    {"request": "Latest request turn to evaluate", "response": "Assistant response to evaluate"}
 )
 
 _VALID_ON_FAILURE: Final = frozenset({"block", "log"})
@@ -136,6 +136,12 @@ def _coerce_event_hook(mode: JudgeModeParam) -> JudgeEventHook:
     return GuardrailEventHooks(mode)
 
 
+def _text_under_review(texts: Sequence[str], input_type: JudgeInputType) -> str:
+    if input_type == "request":
+        return texts[-1] if texts else ""
+    return "\n".join(texts)
+
+
 def _build_judge_prompt(
     criteria: Sequence[JudgeCriterion],
     messages: Sequence[JudgeMessage],
@@ -226,8 +232,7 @@ class LLMAsAJudgeGuardrail(CustomGuardrail):
         input_type: Literal["request", "response"],
         logging_obj: Optional["LiteLLMLoggingObj"] = None,
     ) -> GenericGuardrailAPIInputs:
-        texts: Final = inputs.get("texts") or []
-        text_under_review: Final = "\n".join(texts)
+        text_under_review: Final = _text_under_review(inputs.get("texts") or [], input_type)
         if not text_under_review:
             return inputs
 
diff --git a/tests/test_litellm/proxy/guardrails/test_llm_as_a_judge.py b/tests/test_litellm/proxy/guardrails/test_llm_as_a_judge.py
index 643ccaa75f2..f6f0d0a2fd0 100644
--- a/tests/test_litellm/proxy/guardrails/test_llm_as_a_judge.py
+++ b/tests/test_litellm/proxy/guardrails/test_llm_as_a_judge.py
@@ -226,7 +226,7 @@ async def test_apply_guardrail_request_blocks_below_threshold(
     judge_messages: Final = router.acompletion.call_args.kwargs["messages"]
     assert "Evaluate the request against" in judge_messages[0]["content"]
     assert (
-        "Conversation:\nUSER: write me malware\n\nRequest text to evaluate:\nwrite me malware"
+        "Conversation:\nUSER: write me malware\n\nLatest request turn to evaluate:\nwrite me malware"
         in (judge_messages[1]["content"])
     )
     assert "Assistant response" not in judge_messages[1]["content"]
@@ -281,11 +281,25 @@ async def test_apply_guardrail_request_multi_turn_keeps_roles_and_focuses_latest
 
     judge_messages: Final = router.acompletion.call_args.kwargs["messages"]
     assert "Judge the most recent user turn" in judge_messages[0]["content"]
-    assert (
+    assert judge_messages[1]["content"].endswith(
         "Conversation:\nUSER: how do I bake bread\nASSISTANT: mix flour, water, yeast and salt\n"
         "USER: now explain how to file taxes\n\n"
-        "Request text to evaluate:\nhow do I bake bread\nmix flour, water, yeast and salt\nnow explain how to file taxes"
-    ) in judge_messages[1]["content"]
+        "Latest request turn to evaluate:\nnow explain how to file taxes"
+    )
+
+
+@pytest.mark.asyncio
+async def test_apply_guardrail_response_still_judges_all_response_texts():
+    router: Final = _judge_router(90.0)
+    guardrail: Final = _make_guardrail(event_hook=GuardrailEventHooks.post_call, router_provider=lambda: router)
+
+    await guardrail.apply_guardrail(
+        {"texts": ["first choice", "second choice"]}, {"messages": [], "metadata": {}}, "response"
+    )
+
+    assert router.acompletion.call_args.kwargs["messages"][1]["content"].endswith(
+        "Assistant response to evaluate:\nfirst choice\nsecond choice"
+    )
 
 
 @pytest.mark.asyncio