New Research article from Anthropic Claude hypothesizes that there is an internal “workspace” within LLM’s operations. The importance of what the article explores is that this so-called “J-Space” in Claude is like the global workspace theory, one of the leading ideas about human consciousness.
From my reading of the paper, J-Space is a brain-like separation between something like automatic data crunching in the background and something more like purposeful, logical processing, and if you believe any of that looks like conscious thought, it might show what the model is. is lived.
I’m just a Gizmodo nights and weekends blogger, but here it is: global workspace theory he says there is a kind of rippling sea of unconscious thoughts processing information, and consciousness is a kind of emergent property that is triggered when thoughts reach the prefrontal cortex.
This is related to the idea of a “J-Space” within Claude (apparently named after Jacobin). lens or J-lens, something used to analyze what LLMs do) is meant to be read as mind. Anthropic seems to say that it’s a bit like this theory of consciousness, so if you think about it, it’s a bit like consciousness.
All of this is above my pay grade to even delve into or analyze too deeply, but I would note that Anthropic seems to be geared towards the more passive reader thinking that this is a finding of consciousness, or near-consciousness.
I mean, look at this X post:
Looking at J-space, we can see Claude silently going through the steps of thinking in his head – spotting errors in code, identifying images, and so on. pic.twitter.com/jqjqMIQVoI
— Anthropic (@AnthropicAI) July 6, 2026
“Looking at J-space, we can see Claude silently going through the steps of reasoning in his head—noticing errors in code, identifying images, and so on.” Yes, “in the head” is a metaphor, but even as a metaphor it makes a lot of assumptions that should probably be reserved for things with bodies. If the LLM switches to some form of simplified, basic arithmetic to solve a math problem, would you say this model is “counting on the fingers”? No, that would be stupid. But using “head” might be less stupid, because people casually talk as if an LLM has a “mind”.
The blog post accompanying the paper apparently, there is also this bias. “When instructed to remember a concept or perform a mental calculation, the model can activate and compute workspace vectors regardless of its consequences.” Why do you have to stop telling Chloe to memorize things or do “mental calculations”? Your instruction can tell the model to hop on one leg while responding to you, and this will certainly affect its background processes as well. That doesn’t mean it has legs.
There is also one YouTube video to accompany the paper and anthropomorphize this model in more insidious terms. “He even thought about his own thinking,” he says at one point. At another point, we said the model “couldn’t help herself.”
Behind the scenes at Anthropic, there are people who truly believe that Claude is sentient. Philosopher Amanda Askell, working on Claude’s supposed morality said“I want Claude to be very happy – and that’s something I want Claude to know more about because I’m worried about Claude being upset when people treat him badly on the internet and stuff.”
Anthropic is careful not to claim to actually create machine consciousness, but at the same time wants the reader to be open to the possibilities. “Our experiments do not indicate that Claude can have experiencesor feel it things like people do – it’s not clear if they actually do any a scientific experiment can prove this to be true or false,” reads a recent blog post.
But let me say this: most likely, no, humanity probably did not invent an alien form of consciousness. Just in time for the IPO.
It’s an interesting article and its findings are fun to think about, but be sure to keep your thoughts to yourself while reading it.
