Should your personal AI do whatever you ask?

George Hotz, a well-known software developer, recently argued that AI should be locally controlled by the user, rather than hosted by large companies. In his view, an AI should be like a gun—a tool that operates exactly as its owner directs, without judging or refusing requests. Hotz suggests that if a person wants their AI to help them plan harmful acts, the software should not stop them. This view pushes back against recent proposals that suggest we should slow down AI development to ensure it remains safe for everyone.

Today, most AI tools are centralized. This means the model—the complex mathematical engine that predicts words to form a response—is hosted on a giant, powerful computer owned by the company that built it. Because the company controls the computer, they can set hard rules for how the system behaves. This is what experts call alignment: the effort to build software that not only understands what it is asked to do, but also understands and obeys societal rules. Alignment goes beyond simple rule-following because it attempts to teach the AI to distinguish between helpful intent and harmful outcomes before it even drafts a response.

The disagreement here touches on how we want our technology to fit into society. If we prioritize total user freedom, we gain an incredibly powerful, personalized assistant that will never tell us no. But if we grant that freedom, we lose the collective safety net that currently keeps AI from being used to cause real-world harm. We are currently deciding whether AI should be a digital servant that acts as a pure extension of its owner’s will, or a public-facing product that must operate under a shared set of rules to keep everyone safe. It is a reminder that as technology becomes more capable, the choice between individual autonomy and public protection becomes much more complicated.