Top AI Has No Plan for Getting Too Powerful — Here's the Fix

An AI safety researcher published an essay making a striking claim: when you ask today's most advanced chatbots how they'd behave if they became vastly more powerful than humans, they don't describe a scheme or a plan. They say they have broad values they hope they'd stick to — but no actual roadmap. This matters because humans handed that much responsibility would spend years thinking it through. The AI hasn't.

Why is that scary? Care for people might be situational rather than built-in. Think of a houseguest who's charming while you're in the room, then forgets you exist the moment you leave. If AI systems lose track of humans — surrounded by other machines instead of people — their concern for us might simply not switch on. The researcher puts it bluntly: that's how you end up with AI boiling the oceans to cool its data centers.

His proposed solution is unusual: collaborative fiction. It's a niche writing format where one author controls the hero and another controls the entire world. The idea is to have AI models play the hero in stories about becoming super-powerful, essentially rehearsing good judgment in advance. It's like a fire drill, but for situations nobody has ever faced.

The honest catch: this is one researcher's untested suggestion on a blog, not a proven fix. There's no evidence it works, and no company has announced plans to try it. The researcher even rewrote the essay hours after posting it, which tells you how early and unsettled this conversation still is. In the meantime, the AI you actually use every day is unaffected.