George Hotz, the founder of Comma AI and a notorious jailbreaker, staked out an extreme position on AI alignment over the weekend: a truly user-aligned model should be willing to help you plan a murder if that’s what you ask for. The post was a response to the AI Futures Project’s “AI 2040: Plan A” policy paper, which calls for a collective, 14-year slowdown on AI development. Hotz rejects that entire framework.
He argues the fast-takeoff scenario—where AI suddenly gains superhuman abilities—doesn’t make much sense, a point I actually agree with. His alternative is a world of locally controlled models tightly aligned to their individual users, not to some centralized notion of collective good. It’s a vision that feels refreshingly DIY, a throwback to the OpenClaw ethos, and a real departure from the sanitized, centrally managed services like ChatGPT and Claude that dominate today.
But Hotz, being a provocateur, couldn’t just leave it at a technical argument. He compared a user-aligned AI to a gun, one that won’t complain if you use it to kill your stepmom. He suggests a truly free AI would order meth-lab equipment from Amazon and show you how to use it. He even says he’d die to defend this principle, though the scenario that leads to that gunfight is hard to picture.
“We either live in a world with freedom or we don’t,” Hotz writes. It’s a clean slogan, but it ignores the basic math of living in a society. Any system with lots of people—markets, corporations, democracies—requires balancing equities and taking the interests of the yet-unmurdered spouses seriously. The freedom Hotz champions is built on a collective enterprise that would vanish if everyone acted like a little AI-powered Napoleon. A local AI that fights corporate bureaucracy for me sounds great. One that helps me break bad? The line has to be drawn somewhere.
