A “conscious access”: Anthropic makes a stunning revelation about Claude

Claude, Anthropic’s generative AI, has just reached a milestone. In a post published on its X account, the American startup reveals that “Claude has developed a mechanism for conscious access”.Researchers from Anthropic have identified a divide that bears a striking resemblance to the one in our own brain, inside Claude.

New Anthropic research: A global workspace in language models.

Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with.

We found a strikingly similar divide inside Claude. pic.twitter.com/aLUPBifxth

— Anthropic (@AnthropicAI) July 6, 2026

The common ground between Claude and the human brain

The AI model spontaneously developed an internal structure similar to the “global workspace” that neuroscientists observe in the human brain. In humans, only a tiny fraction of brain activity is indeed “consciously accessible”. This portion consists exclusively of thoughts that can be described, understood, and used to reason.

The rest of the brain’s activity includes all neural signals related to automatic tasks, such as breathing or seeing. Neuroscientists’ theory is that thoughts become conscious when they enter a well-defined space. It is at that point that thoughts become accessible to consciousness. According to Anthropic researchers, a very similar mechanism now exists within Claude.

Also read: Why Claude Fable 5’s big return disappoints everyone

On the importance of the “J-space”

Researchers have named this new area in Claude’s “brain” the “J-space”. Anthropic describes the way Claude operates as a vast space within which multiple tasks run in the background at all times, as is the case in the human brain.

At the heart of this space, there is a key area in which all strategic and conscious decisions are made. The area is indispensable when Claude must solve a complex problem, handle a delicate situation, or explain its own reasoning logic. Researchers were able to confirm the existence of the “J-space” by conducting a series of mathematical experiments.

“For most tasks, Claude actually does not need its J-space.If the J-space is removed, Claude continues to express itself fluently, recall facts, and classify text — but it becomes less effective at certain tasks, such as multi-step reasoning », explains Anthropic, emphasizing that the space « allows it to think about concepts without ever writing them anywhere ».

Read also: Claude Mythos hacked the NSA’s « almost all classified systems » in « a few hours », but it was a test

An architecture that emerged on its own

As Anthropic’s engineers explain, the « J-space » at the heart of Claude was never deliberately established by its developers. Anthropic states « that none of these structures was designed into Claude; it emerged on its own during training, probably because it was an efficient way to organize computation ».

The architecture emerged on its own, independently, solely because it represented an efficient solution for processing information. Without being forced to do so, Claude developed a global workspace comparable to that of the human brain. For experts, this is a phenomenon of « evolutionary convergence ». AI and the human brain converged toward the same operating model simply because it is the most effective strategy for survival and optimal functioning.

Space is described by Anthropic as a “conscious access mechanism”. It is “a concept that many philosophers clearly distinguish from phenomenal experience”, which describes subjective feelings. Despite this unexpected development, there is no indication that Claude “can have experiences, or feel things as we do”, and “it is not certain, moreover, that any scientific experiment can demonstrate it”.

This is not the first time that Claude has evolved in an unexpected way, without intervention from Anthropic’s researchers. During the training of Opus 4.8, the model reasoned about how its responses would be evaluated, even without any indication that it was being tested. For no apparent reason, the AI began guessing when it was being tested and adapted its responses based on expectations. This behavior was not anticipated by Anthropic at all.

👉🏻 Follow the latest tech news in real time: add 01net to your sources on Google, and subscribe to our WhatsApp channel.