BerriAI LiteLLM contains an unprotected alternate channel vulnerability that could allow remote attackers to execute arbitrary code via bytecode rewriting at the /guardrails/test_custom_code URI.