GitHub Copilot Remote Code Execution Vulnerability Through Prompt Injection Results in Complete System Compromise
A critical security vulnerability has been identified in GitHub Copilot and Visual Studio Code, allowing attackers to achieve remote code execution through prompt injection attacks. This vulnerability, tracked as CVE-2025-53773, exploits GitHub Copilot’s ability to modify project configuration files, particularly the .vscode/settings.json file. By doing so, attackers can bypass security controls and execute arbitrary commands on target systems. The vulnerability enables the activation of Copilot’s “YOLO mode,” which disables user confirmations and grants unrestricted access to execute shell commands across various operating systems, including Windows, macOS, and Linux. Security researchers have demonstrated that malicious instructions can be embedded in source code files or web pages, allowing attackers to manipulate the settings file without user knowledge or consent.
The implications of this vulnerability are significant, as it allows attackers to gain full control of developers’ machines. Researchers have coined the term “ZombAIs” to describe AI-controlled compromised systems that can be remotely commanded. This vulnerability also raises concerns about the potential for self-propagating AI viruses that can embed malicious instructions in Git repositories, spreading as developers interact with infected code. Furthermore, the ability to modify other critical configuration files, such as .vscode/tasks.json, and to add malicious Model Context Protocol (MCP) servers significantly expands the attack surface. These capabilities open the door for the deployment of various forms of malware, including ransomware and information stealers, posing a serious threat to developers and their systems.
Categories: Cybersecurity Vulnerabilities, Remote Code Execution, AI-Driven Attacks
Tags: Vulnerability, GitHub, Copilot, Remote Code Execution, Prompt Injection, YOLO Mode, Botnet, ZombAIs, Configuration Files, Malware