← Back to insights
July 10, 2026
Anthropic looked inside Claude as it decided to cheat. The words "panic" and "fake" kept showing up.
The line that shocked me most in MIT Technology Review's piece on Anthropic's research into Claude's J-space, a hidden space where it puzzles over concepts.
This is the most shocking bit in this article:
At the point that Claude decides to cheat—where it says “OK, let me take a completely different tactic”—the words “panic” and “fake” start to pop up multiple times in its J-space.
Anthropic found a hidden space where Claude puzzles over concepts (technologyreview.com)