Notable article and paper by Anthropic on the so called J-Space:

A global workspace in language models
https://www.anthropic.com/research/global-workspace

Verbalizable Representations Form a Global Workspace in Language Models
https://transformer-circuits.pub/2026/workspace/index.html

It does not come as real surprise after the Introspection paper, noteworthy is that the system itself is aware of the J-Space.

Sabine Hossenfelder defined consciousnesses as "world-model + self-model + ability to infer these", how close are we now?