CVE-2025-1474

What it actually does

An administrator can create a new user account without setting a password. This results in an account that may be accessible without any authentication credentials, bypassing the system's login security controls for that specific user.

Why it matters to the business

This vulnerability introduces a significant security gap, as it allows for the creation of password-less user accounts. If an administrator's account is compromised, an attacker could create these insecure accounts as backdoors for persistent access. This could lead to unauthorized access to sensitive machine learning models, experimental data, and intellectual property, potentially resulting in data breaches and operational disruption.

How an attack unfolds

An attacker who has already gained administrative privileges on an MLflow server can exploit this flaw. The attacker would navigate to the user management function and create a new user. By intentionally leaving the password field blank, the attacker leverages the vulnerability to successfully create the account. This new, password-less account can then be used by the attacker or anyone who knows the username to log in without credentials, gaining access to the platform's resources according to the permissions assigned to that user.

How to fix it

Upgrade MLflow to version 2.19.0 or later. If building from source, ensure the patch from commit `149c9e18aa219bc47e86b432e130e467a36f4a17` is applied. After upgrading, it is critical to perform an audit of all existing user accounts to identify any that were created without a password. For any such accounts found, either set a strong, unique password or disable the account immediately.

How to confirm the fix worked

To confirm the vulnerability is remediated, upgrade to a patched version and attempt to create a new user without specifying a password. The system should now correctly reject this action with an error message. To detect existing exposure, administrators should query the user database or use the MLflow administrative interface to inspect all user accounts, specifically looking for any with null or empty password fields.

Typical attack path

1. An attacker first compromises an account with administrative privileges on the target MLflow instance.
2. The attacker uses their administrative access to create a new user account via the API or user interface.
3. During the creation process, the attacker intentionally omits the password for the new user.
4. The vulnerable system incorrectly processes the request and creates the user account without a password.
5. The attacker can now log in as the new user by providing only the username, establishing a persistent and easily accessible backdoor.

How to detect it

Review MLflow audit logs for user creation events, paying close attention to any initiated by administrators. Suspicious activity would involve the creation of new users where the request parameters show a null or empty password field. The most definitive method of detection is to perform a direct audit of the user account database or configuration to find any user entries that lack an associated password hash or credential.

What your security team should do

• Immediately audit all MLflow user accounts to identify and remediate any password-less accounts by setting a strong password or disabling them.
• Prioritize upgrading all MLflow instances to version 2.19.0 or newer to prevent this vulnerability from being exploited.
• Review administrative logs for any unusual user creation events, as this could indicate that an administrator account has been compromised and used to create backdoor accounts.