Anthropic Finds a Conscious-Like Layer in Claude
New Anthropic research identifies a "global workspace" inside Claude — a small slice of active representations that the model can reason with, mirroring how only a fraction of human brain activity reaches conscious awareness. The team calls it "J-space" and says it can be read, audited, and shaped to keep models trustworthy as they grow more capable. They've also partnered with Neuronpedia on an interactive demo of the methods on open-weights models.
This briefing is part of the archive.
Get the app for full archive access — every daily briefing, sourced and attributed.