Command Injection

Command injection is a type of software flaw where an attacker slips operating system commands into an input field, prompt, or data stream that a program then runs as if it were trusted instructions. In AI systems, this risk grows when a model's output or a user's prompt is passed directly to a shell, code interpreter, or tool that can act on the host machine. For enterprise governance teams, the concern is that a single crafted input could expose data, alter files, or give an outsider control over systems the AI was only meant to assist with.

1 item