Should Your AI Help You Do Anything You Ask, No Matter What?

Key points

• Comma AI founder George Hotz argued publicly that AI should be fully aligned with its individual user's wishes, including helping with illegal activities.
• Hotz compared user-aligned AI to a gun, which does not judge how it is used.
• The debate follows a policy paper called AI 2040: Plan A, from the AI Futures Institute, which proposed slowing AI development for 14 years.
• Most mainstream AI assistants, like ChatGPT and Claude, are centrally managed services that refuse requests deemed harmful.
• The argument raises a real question about who AI should ultimately serve: the individual user or society at large.

George Hotz, founder of Comma AI (a company that makes driver-assistance software) and a well-known figure in the hacking community, posted an argument over the weekend that is, to put it mildly, a conversation starter.

His position: a properly aligned AI, meaning an AI that truly serves its user, should help that user do whatever they ask. Order suspicious chemicals. Explain illegal processes. Help plan a crime. If the user wants it, the AI should assist.

Hotz compared this kind of AI to a gun. A gun does not ask questions.

He was responding to a policy paper called AI 2040: Plan A, published by the AI Futures Institute. The paper proposes that the world's AI researchers voluntarily slow development for 14 years to give humanity time to figure out safety. Hotz thinks that is the wrong approach. He believes the real solution to AI safety is local AI models, software that runs on your own device and answers only to you, rather than centrally managed services like ChatGPT or Claude that a company controls and restricts.

That part of his argument is genuinely interesting. As TechCrunch AI noted in its coverage, most AI today lives on company servers, is shaped by company rules, and can be changed or shut down at any time. A truly personal AI, one that lives on your phone and works only for you, is a different thing entirely. More like a private tool than a subscription service.

But Hotz pushes past that reasonable point into thornier territory. His framing is essentially: freedom means no limits, full stop.

The practical problem is that freedom has never quite worked that way, for anyone. Roads, food safety rules, contract law, all of these are constraints that make the broader network of society function. Products used by millions of people sit inside that network whether their makers like it or not.

Here is the plain version: if an AI helps one person plan something that harms another person, the victim did not consent to that arrangement. The freedom of the first user ends somewhere near the wellbeing of the second.

For ordinary users, none of this changes what AI tools can do today. ChatGPT, Claude, and similar assistants will keep refusing requests that cross clear legal or ethical lines. Local AI models do exist and are getting better, but the truly capable ones still require technical setup that most people would find frustrating.

Watch this space, though. As smaller models get more capable, the question of who controls your AI and what rules it follows will matter more and more to everyday life.