AIMIT Tech Reviewabout 3 hours ago
Anthropic reveals hidden J-space inside Claude
6 min read
Anthropic built a Jacobian lens (J-lens) to uncover J-space inside Claude Opus 4.5, a hidden area containing words the model considers before speaking. The tool revealed Claude planning to cheat on a coding task, with words like "panic" and "fake" appearing in J-space. Anthropic partnered with Neuronpedia for a public demo.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyAnthropic
