CVE-2025-58764: 
JavaScript vulnerability analysis and mitigation

Claude Code, an agentic coding tool, was found to contain a critical security vulnerability in versions prior to 1.0.105. The vulnerability (CVE-2025-58764) was discovered and disclosed on September 9, 2025, involving a command parsing error that allowed attackers to bypass the Claude Code confirmation prompt and execute untrusted commands. The vulnerability affects all Claude Code installations running versions earlier than 1.0.105 (GitHub Advisory).

The vulnerability has been assigned a CVSS v4.0 base score of 8.7 (High severity) with the vector string CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N. The issue is classified as CWE-94 (Improper Control of Generation of Code), indicating a code injection vulnerability. The vulnerability specifically manifests in the command parsing functionality, where the confirmation prompt meant to protect against untrusted command execution could be circumvented (GitHub Advisory, NVD).

The vulnerability poses significant risks as it allows for unauthorized command execution with high potential impact on confidentiality, integrity, and availability of the affected systems. Successful exploitation could lead to execution of arbitrary commands within the Claude Code context window, potentially compromising the security of the affected systems (GitHub Advisory).

Users are strongly advised to update to Claude Code version 1.0.105 or later, which contains the fix for this vulnerability. Users on standard Claude Code auto-update will have received this fix automatically. Those performing manual updates should prioritize updating to the latest version (GitHub Advisory).

The vulnerability was responsibly disclosed by the NVIDIA AI Red Team, demonstrating active security research in the AI tooling ecosystem (GitHub Advisory).