Anthropic Taps Claude to Help Build Its Successor, Raising Fresh Self‑Improvement Questions

U.S. AI lab Anthropic announced on September 17 that its large language model Claude is now deeply involved in developing the next, more capable generation of Claude. The company is using the current model to help with code generation, experiment design, and evaluation, aiming to speed up work on its flagship “frontier” systems.

According to Anthropic, Claude assists researchers under human direction with tasks such as preprocessing training data, comparing new architecture variants, and interpreting benchmark results. Internally, it is treated less like a passive tool and more like a “co‑developer” working alongside human engineers. That approach brings efficiency gains, but it also nudges the field toward a controversial idea: AI systems that help build more powerful versions of themselves.

Self‑improvement has long been cited in debates over runaway or hard‑to‑control AI. Anthropic CEO Dario Amodei has recently joined others in calling for a slowdown of frontier‑model scaling and for stronger transparency around progress. In parallel with the announcement, the company published proposed metrics for tracking capability growth inside major labs, inviting oversight from governments and independent researchers. The big question now is whether such safeguards can keep pace with increasingly self‑directed development workflows.