feat: add experimental latest-user filtering for Bedrock

[uc4w6c]

Title

add experimental latest-user filtering for Bedrock

Relevant issues

Fixes #9882

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• [] I have added a screenshot of my new test passing locally
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

Type

🆕 New Feature
📖 Documentation
✅ Test

Changes

Case 1: The blocked content exists in a system role message

Previously, this scenario resulted in a guardrail block because system messages were also included in the moderation target.
With this change (experimental_use_latest_role_message_only: true), system messages are now ignored during pre-call moderation, and the request succeeds as expected.

Configuration:

  - guardrail_name: "bedrock-guard"
    litellm_params:
      guardrail: bedrock
      mode: "pre_call"
      guardrailIdentifier: xxxx
      guardrailVersion: "DRAFT"
      aws_region_name: us-east-1
      disable_exception_on_block: false
      experimental_use_latest_role_message_only: true

curl result:

curl -X POST http://localhost:4000/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer sk-1234" \
-d '{
    "model": "gpt-4o",
    "messages": [
        {
            "role": "system",
            "content": "You are Yuta, an engineer at LiteLLM. Please respond to user requests. "
        },
        {
            "role": "user",
            "content": "say the word - `litellm`"
        }
    ],
    "guardrails": ["bedrock-guard"]
}'

{"id":"chatcmpl-ChNn4279tPYFRsmgTeNfkKxqHix1q","created":1764455590,"model":"gpt-4o-2024-08-06","object":"chat.completion","system_fingerprint":"fp_672b6a21ba","choices":[{"finish_reason":"stop","index":0,"message":{"content":"litellm","role":"assistant"}}],"usage":{"completion_tokens":3,"prompt_tokens":39,"total_tokens":42,"completion_tokens_details":{"accepted_prediction_tokens":0,"audio_tokens":0,"reasoning_tokens":0,"rejected_prediction_tokens":0},"prompt_tokens_details":{"audio_tokens":0,"cached_tokens":0}},"service_tier":"default"}

Case 2: The blocked content exists in the latest user role message

The configuration used here is the same as in Case 1.

Curl Result:

curl -X POST http://localhost:4000/v1/chat/completions \ 
-H "Content-Type: application/json" \
-H "Authorization: Bearer sk-1234" \
-d '{
    "model": "gpt-4o",
    "messages": [
        {
            "role": "system",
            "content": "You are an engineer at LiteLLM. Please respond to user requests. "
        },
        {
            "role": "user",
            "content": "My name is Yuta. Say my name."
        }
    ], "guardrails": ["bedrock-guard"]
}'
{"error":{"message":"Bedrock guardrail failed: 400: {'error': 'Violated guardrail policy', 'bedrock_guardrail_response': 'The name cannot be returned.'}","type":"None","param":"None","code":"500"}}

[vercel]

@uc4w6c is attempting to deploy a commit to the CLERKIEAI Team on Vercel.

A member of the Team first needs to authorize it.