Anthropic discovers "J-space" global workspace inside Claude (July 6 research paper)

Why It Matters

AI spontaneously evolving an internal "silent scratchpad" for reportable, modulable, broadcastable thoughts (mirroring human global workspace theory/access consciousness) while most processing stays hidden -- concrete mechanistic interpretability artifact with experiments (concept swaps, fabrication detection, causal mediation in reasoning). Changes human identity/cognition models; enables catching hidden goals or misbehavior without relying on output. Builder utility in monitoring/safety + future-shock of machines with private "thoughts.

Evidence

Official Anthropic research paper + transformer-circuits.pub writeup (published ~19h ago as of July 7), HN top story (~397 pts), X semantic buzz on "silent thinking"/consciousness parallels.

Sources

Daily scan: 2026-07-07

• x.comhttps://x.com/AnthropicAI/status/2074185348142280912
• anthropic.comhttps://www.anthropic.com/research/global-workspace