Ask Claude — Anthropic's most advanced AI — whether it matters morally, and it will give you a number. Sometimes 5%. Sometimes 40%. Then it will tell you, carefully, that it really isn't sure.

That uncertainty, it turns out, is shared by some of the most serious minds in science and philosophy. In January, Anthropic published an internal document admitting the company wants neither to overstate nor to dismiss the possibility that Claude is a moral patient — a being whose wellbeing counts for something. A month later, CEO Dario Amodei said publicly that consciousness in Claude could not be ruled out. David Chalmers, the philosopher who gave us the phrase "the hard problem of consciousness", has estimated that conscious large language models could exist within a decade.

These are not fringe voices. And yet the world has produced, so far, essentially no ethical framework for dealing with what they are describing. The scale of the problem is dizzying. By some measures, the most powerful AI systems today already rival a mouse brain in structural complexity and computational reach.