1[medium] mlflow: MLflow CrewAI autologging records LLM provider API keys in logged trace spansgithub.comHigh10·27 21:26Z
MLflow disclosed a medium-severity vulnerability (CVE-2026-116691) in which CrewAI autologging recorded LLM provider API keys in plaintext within trace spans1. The flaw, affecting MLflow versions 2.19.0 through 3.16.0, stems from the _get_agent_attributes() function serializing CrewAI Agent objects without sanitizing credential-bearing fields such as llm and function_calling_llm. Testing on MLflow 3.16.0 with CrewAI 1.15.18 found the exposed credential in five attribute locations across two spans. MLflow 3.17.0 patches the issue.