A critical vulnerability, CVE-2026-0830, was discovered in AWS Kiro, an AI-powered IDE extension. This was not a sophisticated new AI hack, but a classic Command Injection bug that led directly to Remote Code Execution (RCE). The flaw existed because the system used a dangerous Node.js method to execute Git commands, creating a massive security risk for developers trusting their local environment.

The technical root cause was a fatal choice in the code: using `child_process.exec` instead of `child_process.spawn`. The vulnerable function constructed a command string using direct string interpolation for the working directory. Because this variable was not sanitized, an attacker could inject malicious shell commands. This allowed for a devastating exploit triggered simply by opening a repository with a specially crafted directory name.

An attacker just needed to convince a developer to clone a folder with a name containing malicious code. When the AI agent queried Git status, it would execute the hidden command with the developer's full privileges, potentially stealing SSH keys or AWS credentials. The fix in version 0.6.18 was architectural, moving to `child_process.spawn` with a `cwd` option, which tells the OS to handle the path safely without a shell.