MLflow versions from 2.19.0 up to but not including 3.17.0 have a vulnerability in the CrewAI autologging feature where API keys for large language model providers are recorded in plaintext within logged trace spans. This occurs because the autolog function converts Agent objects to strings including sensitive credentials and expands these into trace inputs, potentially exposing provider API keys to anyone with access to the trace backend or OTLP destination. The issue is fixed in version 3.17.0.