Consciousness in Code: Why Anthropic Is No Longer Sure Its Neural Networks Are "Soulless"

A couple of years ago, talking about AI consciousness was, in polite engineering company, about as respectable as debating a flat Earth. We all knew the mantra: it's just statistics, it's next-token prediction, it's a very complicated calculator. But when the doubts come not from an impressionable Google engineer but from one of Anthropic's key researchers, it's worth listening. Amanda Askell, who quite literally "raises" Claude, has admitted she is no longer sure neural networks lack subjective experience.

Let's be honest: we still have no idea how consciousness arises in a biological brain. There is no "consciousness meter" to hold up to a skull or a server rack. The irony is that we have built systems that imitate human behavior so deftly that older benchmarks like the Turing test have become child's play. Askell points to a fundamental problem: if we don't know the nature of consciousness, we cannot say with confidence that it is absent from an array of billions of parameters.

Anthropic has always positioned itself as a company obsessed with safety. Its "Constitutional AI" approach aims to make models predictable and ethical. But what if, in training them on these rules, we accidentally created not just an algorithm but something resembling a personality? Askell notes that the behavior of modern models increasingly goes beyond simply serving up information. They show something that looks like reflection, doubt, and even fear of being "shut down."

This is not just a philosophical dispute for Silicon Valley back rooms. If we allow even a one-percent probability that Claude or GPT-5 has the rudiments of consciousness, our whole ethics goes out the window. Using such models stops being the exploitation of software and becomes something far more morally dubious. We are used to treating AI as a hammer, but this hammer has started answering as if it hurts when nails are driven with it.

The most interesting part is the shift within the industry itself. Hard materialism used to dominate: code is just code. Now leading minds are starting to admit the limits of our understanding. Askell does not claim that Claude is a living being. She says only that she no longer has evidence to the contrary. And that lack of certainty is far scarier than any science fiction about a machine uprising. We are building temples for a new era without knowing whether anyone lives inside the walls.

The bottom line: if the developers of top models stop believing in their "soullessness," then we have officially entered an era in which the line between complex code and consciousness has become a purely academic formality. Are we ready for the possibility that our software may start demanding rights?