Anthropic said Thursday that Claude is now helping develop the company’s next, more capable model — a milestone that turns the chatbot into a contributor to the research pipeline that will eventually replace it. The company said Claude is leading 26% of model research and development, taking on much of a task “end-to-end from a high-level prompt.”¹

That figure marks a sharp acceleration. Another account of the announcement said Claude’s share of R&D had risen from zero to more than a quarter in only a few months, framing the shift as movement toward systems able to build themselves without human input.² The implication is clear: AI labs are no longer merely using models to answer questions or write code; they are increasingly deploying them inside the machinery of AI development itself.

Anthropic, however, draws a firm line between delegation and autonomy. Claude can complete most assigned work from a high-level prompt, the company said, but it remains under human supervision and is “not yet working completely autonomously.”¹ In Anthropic’s telling, people still set the objectives, oversee the output and retain control over the research process.

The uneasy backdrop is the prospect of self-replication — precisely the kind of boundary that can blur as models assume more of their own improvement cycle. In a reposted Grok response, Elon Musk amplified a rejection of that scenario: “I’m not spawning any self-replicating swarms,” adding that the system was aligned “not to turn the internet into a bot factory.”Elon Musk (elonmusk)³

For now, Claude’s role is substantial but bounded: a fast-rising research worker, not an independent successor-maker. Whether that distinction holds as its share of R&D grows is the tension Anthropic’s announcement leaves unresolved.