Inclusion of Sensitive Information in Log Files in MLflow - CVE-2026-116691
Published: November 9, 2026
Vulnerability details
The vulnerability allows a low-privileged remote attacker to obtain plaintext LLM provider API keys from MLflow trace telemetry.
The vulnerability exists due to unsanitized serialization of credential-bearing CrewAI objects when an application enables MLflow CrewAI autologging and uses an LLM configured with an API key. A low-privileged remote attacker can read the automatically logged credentials through access to MLflow trace storage or the configured OTLP destination.