How a Malicious git.exe Could Run Through Claude Desktop on Windows Opening a software project in Claude Desktop on Windows could trigger malicious code without the user deliberately running it. Anthropic disclosed a security vulnerability, CVE-2026-88066, on October 13, 2026. The flaw affected Claude Desktop versions before 1.8089.0 and was rated High severity, with a CVSS 4.0 score of 7.5. The problem involved how Claude Desktop launched Git, a tool commonly used to manage software projects. Instead of specifying the full path to a trusted Git executable, the application invoked git.exe by name. Under the affected conditions, Windows could find a malicious file named git.exe inside the opened project directory and execute it instead of the legitimate program. An attacker could exploit this behavior by placing a malicious executable inside a repository that another person would later open. The distinction matters: this was not a case of Claude's AI model deciding to execute harmful instructions. It was a software vulnerability involving executable search paths and operating-system behavior. Anthropic fixed the issue in Claude Desktop version 1.8089.0 and reported distributing the update through its normal automatic-update mechanism. The vulnerability required the user to open an affected project, but not to manually launch the malicious executable. Publicly verified evidence of attacks exploiting this specific flaw has not been established. The incident illustrates why AI coding tools must secure not only their AI-generated actions, but also the ordinary development utilities they invoke. #ClaudeAI #Cybersecurity