FrontierThe story, in brief
Anthropic found a hidden space where Claude puzzles over concepts
Anthropic just cracked open Claude's black box. Here's what's actually happening inside.
The KeyNews take
Why it matters
Anthropic's Jacobian lens technique offers unprecedented interpretability into how LLMs process reasoning—a capability breakthrough that could reshape how teams evaluate and trust AI systems in production.
The key facts
5 to knowAnthropic developed 'Jacobian lens' tool for LLM interpretability
Technique reveals internal reasoning patterns in Claude
Findings range from 'mundane to unnerving'
First clear glimpse into model internals during task execution
Published July 9, 2026 in MIT Technology Review
Go to the source
MIT Technology Review AItechnologyreview.com
Publisher excerpt: The AI firm Anthropic has developed a technique that has given it the clearest glimpse yet at what’s really going on inside large language models as they answer questions or carry out tasks. What they found ranges from the mundane to the unnerving. Researchers at the company built a tool called the…