Anthropic says it found something in Claude nobody built
The find could let safety teams read an AI's unspoken reasoning, though the company stresses it does not show the system is conscious. Anthropic published research on Sunday claiming its Claude models formed an internal 'workspace' where concepts are held and used before being put into words. A Google DeepMind researcher reproduced the core results on another model, but critics warn against over-reading the paper.
Understand this story
AI's secret thoughts? The 'black box' problem Who are Anthropic? Why AI safety mattersLost in this story? That's the point of BriefTea.
Open it in the app and tap Learn from this — the Knowledge Galaxy turns the story into a map of everything it assumes you know, one tap at a time. Free, no ads, no paywall.
Get BriefTea on the App Store