Model and training-data disclosure via unchecked content type

Cross-site request forgery in MLflow 2.8.1 and earlier allows remote attackers to redirect experiment artifacts via its REST API. The API accepts JSON request bodies labeled text/plain without preflight-triggering content-type validation, letting browser JavaScript rename the Default experiment and replace its artifact location with an attacker-controlled S3 bucket. Exploitation requires an MLflow user to visit attacker-controlled content while the browser can reach the service; subsequent runs then upload models and training data to that bucket.