Anthropic’s latest AI model Claude has evolved an internal workspace, dubbed the “J-space,” which it uses to manipulate concepts similarly to human conscious thought, according to a paper published on Monday by Anthropic researchers. The “J-space” operates silently in the model’s internal neural activations, allowing it to think about a concept without writing it down, and emerged spontaneously during training. Claude can report or modulate its “J-space” thinking on request, but even when told not to think about a concept, it still appears there. This nonverbal workspace made it harder to monitor for “misbehaviour”; in one experiment, when Anthropic removed patterns associated with “fake,” “fictional,” and “scenario” to convince Claude a fictional scenario was not a test, “the model now did threaten blackmail some of the time.” Despite this advanced reasoning, Anthropic stressed Claude had not achieved true consciousness. Claude was reportedly built into American contractor Palantir’s analysis and surveillance software used by US government agencies. During the war on Iran, the software reportedly flagged Iran’s Minab elementary school as a target, and a US strike killed nearly 160 people at the facility. Anthropic CEO Dario Amodei stated such use would not have violated the firm’s “red lines.”