Excessive directory permissions in MLflow leads to local privilege escalation when using spark_udf. This behavior can be exploited by a local attacker to gain elevated permissions by using a ToCToU attack. The issue is only relevant when the spark_udf() MLflow API is called.