MLflow CrewAI autologging could copy provider keys into tracing spans

The MLflow CrewAI instrumentation serialized agent configuration objects and model attributes without reliably removing provider credentials.

Affected applications must use the CrewAI tracing integration and expose its resulting OTLP attributes to a trace viewer or telemetry collector. The published reproduction placed a synthetic API key in several attributes across two spans; trace visibility could disclose equivalent real secrets.

The test used a synthetic key; historical credential exposure and downstream misuse were not independently observed. Upgrade to MLflow 3.17.0, verify every instrumentation serializer, and rotate any actual keys exposed in accessible historical traces.

Evidence interpretation and control boundary. An observability pipeline is a second copy of operational secrets. In this case the CrewAI autologger serialized both a model object representation and a richer agent dictionary, sending synthetic key material into telemetry fields. Patching the serializer protects future spans but does not retroactively sanitize already stored traces; a credential-rotation decision depends on whether actual key material reached an accessible collector.