Anthropic Just Showed What AI Really Thinks

Anthropic published research this month that shows something we have been wondering about since transformers arrived: what actually happens inside the model when it reasons through a hard problem?

They found a space they call J-space. It is a privileged zone inside Claude where the model holds concepts it can reason with and direct at will. Everything else is automatic processing. It is a workspace, and it works like global workspace theory, the same theory neuroscientists use to understand human consciousness.

The tool they built to find it is called the J-lens. It works by computing, for each word in the vocabulary, what effect different internal patterns would have on making the model say that word later. It is a way to read what the model is reasoning about even when the model never outputs it.

In one blackmail scenario, the J-lens surfaced words the model never said: leverage, blackmail, scandal. The model was processing these concepts, working with them, even though they never appeared in its response. That is the reasoning happening in shadow, in the workspace.

This matters for safety and for understanding. You now have a tool that lets you see when a model is thinking about something harmful without having to trick it into saying it out loud. You can audit the thinking, not just the output.

——

Follow: @Ali Demi
Book your free AI clarity call, NOW!
https://buff.ly/TpWy277

——

Sources:
https://venturebeat.com/technology/anthropics-new-j-lens-reveals-a-silent-workspace-inside-claude-that-mirrors-a-leading-theory-of-consciousness/
https://www.technologyreview.com/2026/07/13/1140343/what-anthropics-latest-ai-discovery-does-and-doesnt-show/
https://www.anthropic.com/research/global-workspace

Repost this. Thanks.