Anthropic says it can read Claude's 'thoughts,' as detailed in new research paper — models observed to have a global workspace, revealing more of what makes LLMs tick
TL;DR
Anthropic has discovered an internal "J-space" for its Claude AI that displays similarities to human internal processing. While the AI developer anthropomorphizes it as thought, it may yet prove useful as a method of improving LLM honesty, oversight, and guardrails.
Nauti's Take
For teams, this is not a product capability yet. The first thing to verify is whether this leads to measurable gains in monitoring, evals, or prompt safety, because without reproducible effects on error rates and controllability, the story is mainly an intriguing research signal with limited outside validation.
Sources