Ultralytics torch_safe_load Vulnerability Allows Remote Code Execution
A command injection vulnerability in Ultralytics' torch_safe_load library enables remote attackers to execute arbitrary code on affected systems.

A critical command injection vulnerability has been identified in the Ultralytics torch_safe_load library, allowing remote attackers to execute arbitrary code on vulnerable systems. The Zero Day Initiative (ZDI) has assigned this vulnerability a CVSS score of 7.8, underscoring its significant risk.
The specific flaw lies within the torch_safe_load function itself. Researchers discovered that the library fails to properly validate user-supplied input before incorporating it into system commands. This oversight permits an attacker to inject malicious commands that are then executed with the privileges of the current user, potentially leading to a full system compromise.
Exploitation of this vulnerability requires a degree of user interaction. Attackers must trick a victim into visiting a malicious webpage or opening a specially crafted file that triggers the vulnerable function. This social engineering aspect means that while the vulnerability is severe, it is not directly exploitable without some form of user engagement.
The disclosure timeline indicates that ZDI first reported the vulnerability to the vendor on June 18, 2026. After follow-up communications and reminders, ZDI ultimately decided to publish the advisory as a zero-day on October 28, 2026, due to a lack of vendor response or patch.
Given the nature of the vulnerability, which involves command injection through user-supplied data, the primary mitigation strategy recommended is to restrict interaction with the product. This means users should be cautious about opening files or visiting links from untrusted sources that might interact with the torch_safe_load functionality.
While no specific CVE ID has been publicly assigned by MITRE for this particular vulnerability in the advisory, the ZDI advisory number ZDI-26-814 serves as the primary identifier. The vulnerability was discovered by Bongeun Koo, who is credited for their work in identifying and reporting this security flaw.
The lack of a timely vendor response highlights a common challenge in the software supply chain, where dependencies and libraries can introduce significant security risks. Organizations relying on Ultralytics or its components should monitor for any official security advisories or patches that may be released in the future.
This vulnerability serves as a reminder of the importance of secure coding practices, particularly in libraries that handle user input and interact with the operating system. Thorough input validation and sanitization are crucial to prevent such command injection flaws from being exploited.