Anthropic's research team made an unexpected discovery: an internal structure was found within the depths of their language model Claude that was not originally built by the developers. This structure, named "J-space," functions as a kind of shared workspace that various components of the model access.
According to a study published on July 6, J-space is a virtual "board" where Claude pools key information while solving tasks. This space was not designed by engineers—it emerged spontaneously during the model's training. Using a special tool called "J-lens," researchers were able to observe how data moves within the AI during task execution.
Analogy with the human brain
The concept of J-space remarkably echoes what neuroscience calls the "global workspace." In humans, this system provides simultaneous access to important information for multiple thought processes. For example, when we hear a question, recall a relevant fact, and decide how to respond, the brain brings all this data together.
Experiments showed that Claude can not only describe the contents of J-space upon request but also modify it. Moreover, manual intervention in this space directly affected the model's responses and behavior.
Implications for AI safety
This discovery has enormous significance for the safety and interpretability of artificial intelligence. The ability to track J-space activity opens a path to identifying hidden motives in model behavior. Monitoring this space will allow for more effective detection of prompt-injection attacks, where malicious instructions attempt to hijack the model's response control.
Although capabilities are still limited—most information processing still occurs outside J-space—Anthropic has already published the source code for the J-lens implementation and released a demo version on Neuronpedia, inviting the scientific community to verify the results.
It is important to emphasize: the company does not claim that Claude possesses consciousness or subjective experience. The term "consciously accessible" information is borrowed from cognitive science and does not imply the presence of actual consciousness.
Expert opinion: This discovery is not just a technical curiosity. It brings us closer to understanding the "black box" of large language models. If we learn to read AI's thoughts in real time, it will fundamentally change the approach to auditing and verifying artificial intelligence systems, which is critically important for their deployment in the financial sector and other high-risk areas.