A consciousness expert just read Anthropic's Claude research. He is not convinced.
Anthropic published findings hinting that its AI model Claude might show early signs of consciousness. A leading brain scientist says the evidence does not hold up.
Key points
- Anthropic, the AI company behind the Claude chatbot, published research in 2025 suggesting Claude's inner workings may show early signs of consciousness.
- The researchers stopped short of claiming Claude is conscious in the same way humans are.
- Anil Seth, professor of cognitive and computational neuroscience at the University of Sussex and co-director of its Centre for Consciousness Science, publicly disputes the finding.
- Seth argues that producing human-like language does not mean an AI system actually experiences anything.
- The debate shapes how AI companies present their products and how regulators might treat them.
Last week Anthropic, the San Francisco AI company, published new research on Claude, its large language model (the technology that powers chatbots like ChatGPT). The researchers said they found signs of consciousness emerging inside the model's inner workings. They stopped short of saying Claude is conscious the way you or I are, but the paper was enough to set off a fresh round of headlines.
Cue Anil Seth.
Seth is a professor of cognitive and computational neuroscience at the University of Sussex, which basically means he spends his career trying to understand what consciousness actually is and where it comes from. Writing for The Guardian, he says he has real doubts.
His central point is a simple one. A language model is trained to predict the next word in a sequence. It does this extraordinarily well. But doing something well, Seth argues, isn't the same as feeling it. He uses a tidy comparison: a detailed computer simulation of a hurricane moves air around convincingly, but it can't blow your roof off. The simulation isn't a real storm. By the same logic, a model that talks about having feelings isn't necessarily having them.
What did Anthropic actually find? The company's researchers looked at internal signals inside Claude as it processes text, searching for patterns that might echo how emotions or awareness appear in human brains. Seth's concern is that spotting a pattern isn't the same as proving experience. A thermostat responds to temperature. Nobody calls it sentient.
For ordinary users, none of this changes what Claude can do today. You can still use it to draft emails or plan meals. Our earlier story on 13 July found that Anthropic had already discovered words flickering inside Claude that never appear in its answers, a genuine finding, but not a window into a robot mind. On the privacy side, Anthropic says it may use your conversations to train future models unless you opt out in settings, so think before you paste in anything sensitive.
Should users be worried about Claude becoming conscious?
No, not in any practical sense right now. Scientists don't yet agree on what consciousness requires in a biological brain, let alone a software system. Anthropic's research opens questions; it doesn't close them. The thing worth watching isn't whether Claude feels something: it's whether companies start using this kind of research to soften regulation before the science has settled. A very impressive word-predictor is still, when you strip it back, a word-predictor.
