Anthropic's research team has achieved a breakthrough in understanding the internal architecture of large language models. During an analysis of the Claude family of models, a structure was discovered that developers had not originally built into the design. This internal mechanism, dubbed "J-space," functions as a common workspace: various components of the model access it to exchange and process key information.
The research results, published on July 6, represent a significant step forward in interpreting what exactly happens inside the "black box" of modern AI systems.
What is J-space and how does it work?
J-space is an internal virtual space where Claude collects and transmits important data throughout the model. The simplest way to think of it is as a "virtual whiteboard" inside the artificial intelligence. When Claude answers a question, solves a problem, or follows an instruction, key information appears in J-space so that different parts of the model can work with it.
This structure was discovered using a special tool called J-lens. Researchers observed how information moves within the model while performing a task. The key point: J-space emerged spontaneously during the training process — it was not directly built in by the designers.
This concept echoes the "global workspace" theory in neuroscience. In humans, this system provides access to important information for multiple thought processes simultaneously. For example, when a person hears a question, recalls a needed fact, and decides how to answer, the brain brings all necessary data together in one place.
Claude works in a similar way. Anthropic showed that the model can not only describe the contents of J-space upon request but also modify it if asked to do so. Moreover, when researchers manually changed J-space, Claude's answers and its behavior when performing tasks also changed.
Why this matters for AI safety and interpretability
This discovery has enormous implications for artificial intelligence safety. If scientists can track J-space activity, they will have a chance to identify hidden motives in the behavior of AI models. This could make it more effective to detect when a system begins to operate unreliably.
This includes detecting attacks. Monitoring J-space makes it possible to see the moment when a model encounters attempts at prompt-injection (inserting malicious instructions into a query to hijack control of the model's response). Additionally, even partial access to such a "conscious" processing layer opens up important prospects for research.
For now, capabilities are limited. Most of Claude's information processing still occurs outside J-space. Nevertheless, Anthropic has open-sourced the J-lens implementation and posted a demo on Neuronpedia, inviting the research community to verify the results in practice.
The research builds on a series of earlier works. In October 2025, Anthropic published a report on emerging introspective awareness. Earlier, in April 2025, the company launched initiatives to study model well-being, step by step advancing the understanding of what happens inside its systems.
There is also an important caveat. Anthropic emphasizes that it is not claiming Claude possesses consciousness or subjective experience. The work uses the term "consciously accessible" information, borrowed from cognitive science — it does not imply the presence of actual consciousness.
Analyst's comment: The discovery of J-space is not just an academic curiosity. For the AI market and related technologies, it is a signal: the "black box" of neural networks is gradually becoming more transparent. The ability to monitor internal processes directly impacts the security of smart contracts and DeFi protocols that use AI agents. The more we understand a model's internal logic, the more reliably we can integrate it into critical financial systems.