Anthropic probes Claude reasoning as world models draw focus

Anthropic announced that it had found a new window into its models’ “internal thoughts” as they reason through answers, raising fresh questions about what can be learned from its latest research. The work is presented as a closer look at how Claude operates, while leaving open limits on how fully AI systems can be understood from the outside.

Researchers are also focused on world models, a technology aimed at helping AI systems better understand the physical world. Current systems can generate text, images, and code with impressive skill, but still struggle with real-world complexity. MIT Technology Review plans a LinkedIn Live discussion with Will Douglas Heaven and Sam Sinha on how world models could affect robotics and intelligent machines, scheduled for 9:30 PDT, 12:30 PM EDT, and 5:30 PM BST.

Other major technology developments include New York becoming the first state to enact a data center moratorium, with its governor banning large data-center construction for up to a year. Smartphone shipments hit a 13-year low after falling 11% in the second quarter of 2026, while Nvidia halved its Asia buyer list to reduce the risk of AI chips reaching China.

Additional stories cover Russian state hackers targeting routers, sugar molecules found in interstellar space for the first time, a new cell therapy that saved four children with terminal brain cancer, the LAPD halting use of Flock surveillance cameras, and US approval for a space mirror designed to reflect sunlight onto Earth. Anthropic also reported that Claude’s values vary by language, and SoftBank CEO Masayoshi Son predicted AI will overtake human intelligence by 2040.