scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-26 06:03:19 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 151

fix: always prioritize messages for prompt redaction to prevent bypass

Browse source 
- Remove conditional check that could discard message-based redactions
- Always use messages as source of truth after guardrail processing
- Fixes MEDIUM severity: prompt precedence bypass when multiple guardrails apply different masks
This commit is contained in:
Aishwary Dongre 2026-10-15 17:05:41 +05:30
parent 11b31dc37a
commit 58b5df4aa0
1 changed files with 5 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

11
litellm/proxy/image_endpoints/endpoints.py
View file

@ -145,12 +145,11 @@ async def image_generation(
user_api_key_dict=user_api_key_dict, data=data, call_type="image_generation"
)
# Guardrails may redact data["prompt"] directly. Only rebuild from messages if prompt wasn't modified.
prompt_after_hook: Final = data.get("prompt")
if prompt_after_hook == prompt_value:
messages: Final = data.get("messages")
if isinstance(messages, list) and messages:
data["prompt"] = get_str_from_messages(messages)
# Guardrails may redact messages. Always use messages as source of truth if they exist.
# If both prompt and messages were modified, prefer messages since that's what guardrails processed.
messages: Final = data.get("messages")
if isinstance(messages, list) and messages:
data["prompt"] = get_str_from_messages(messages)
data.pop("messages", None)
## ROUTE TO CORRECT ENDPOINT ##