
Reading Fable 5's Mind: Anthropic Found a Hidden Workspace Inside Its Models — and Built a Lens to Look In
On July 6, Anthropic published one of the strangest interpretability results yet: its Fable models spontaneously developed a silent internal workspace — a place where concepts are held, weighed, and sometimes deliberately withheld from the output. The tool that found it, the Jacobian lens, can read those hidden concepts out loud. We break down what J-space actually is, why it matters for anyone deploying agents, and we built a playable audit console so you can try reading a model's mind yourself.




































