Authentication Bypass Vulnerability in FastAPI Job Endpoints of MLflow

CVE-2026-0545

What is CVE-2026-0545?

CVE-2026-0545 is an authentication bypass vulnerability found in the FastAPI job endpoints of the MLflow open-source platform. MLflow is widely used for managing the machine learning lifecycle, including experimentation, reproducibility, and deployment. This vulnerability arises when the basic-auth application is enabled but does not adequately protect the FastAPI job endpoints located under /ajax-api/3.0/jobs/*. Specifically, it allows any network client to execute job functions without proper authentication or authorization when the job execution feature is enabled.

If the vulnerability is exploited, it could lead to unauthorized actions on the part of an attacker, such as submitting, reading, searching, or canceling jobs without the need for valid credentials. This not only constitutes a severe breach of access control but also raises the risk of unauthenticated remote code execution if the allowed job functions can perform high-privileged actions, such as shell execution or changes to the filesystem. Even jobs considered safe can lead to serious issues, including job spamming, denial of service (DoS), or exposure of sensitive data through job results.

Potential impact of CVE-2026-0545

1. Unauthorized Access: The vulnerability allows attackers to bypass authentication measures, enabling them to submit and manipulate jobs within the MLflow environment without any credentials. This could lead to unauthorized access to sensitive data and functionalities.

2. Remote Code Execution: If certain job functions are designed to execute code or manipulate the system, this vulnerability can allow attackers to execute arbitrary code remotely, which can compromise the entire system or lead to data leakage.

3. Denial of Service (DoS): By leveraging this vulnerability, malicious actors could flood the system with excessive job requests, leading to service degradation or outright denial of service for legitimate users, impacting the availability of critical machine learning operations.

References