Critical Vulnerability in MLflow Allows Unauthorized Access to Metadata Services

## Overview MLflow has a critical vulnerability identified as CVE-2026-64849. This issue affects versions prior to 3.15.0. An unauthenticated POST request to the endpoint `/api/2.0/mlflow/webhooks/{id}/test` can be exploited by attackers. ## Technical Details The vulnerability arises because the `_validate_webhook_url()` function in `mlflow/utils/validation.py` only checks the original URL. However, the webhook delivery mechanism in `mlflow/webhooks/delivery.py` follows redirects and re-resolves the hostname. This behavior does not pin the validated address, which creates a pathway for attackers to access internal or cloud metadata services. ## Impact An attacker can exploit this vulnerability to retrieve sensitive information such as `response_status` and `response_body`. This can lead to unauthorized access to critical metadata services, potentially exposing sensitive data and configurations. ## Mitigation Defenders should immediately upgrade to MLflow version 3.15.0 or later. This version addresses the vulnerability by ensuring proper validation of webhook URLs. Regularly check for updates and apply security patches promptly to minimize exposure to such vulnerabilities.