Anthropic’s Claude Takes Lead in Building Its Own Successor

In a blog post that raised a few eyebrows, Anthropic revealed that its flagship model, Claude, is now pulling a surprisingly large part of the heavy lifting behind the next‑generation AI they’re working on. The company says Claude is "leading" roughly 26% of its model research and development, meaning the system can take a high‑level prompt and see the job through to completion—still under human watch‑eyes, of course.

It’s not a full‑blown autonomous creator yet. Anthropic stresses that about 90% of the work is still a partnership: Claude does large chunks, but people are in the loop, guiding and correcting as needed. Think of it as a very smart assistant that can draft the blueprint while a human engineer double‑checks the details.

That collaboration has ramped up fast. Back in February Claude barely touched any of the R&D pipeline. Six months later it was handling a quarter of the work, a jump that the lab calls “significant.” By August the model was leading one‑quarter of research tasks, a milestone the company highlighted as evidence of AI’s growing self‑improvement capacity.

But the excitement is tempered by caution. A chorus of AI leaders—including Anthropic’s own CEO Dario Amodei, OpenAI’s Sam Altman, and even Elon Musk—have been urging a slowdown, warning that letting models like Claude iterate on themselves could make them harder for humans to understand or control. Anthropic’s response? Push for more openness. They want the public to see concrete metrics on how close labs are to achieving what’s called “recursive self‑improvement,” the idea that an AI could design a better version of itself without heavy human input.

To that end, Anthropic is calling on other labs to adopt a shared reporting framework, so the numbers can be compared over time and across companies. They also disclosed that, as of August, roughly 30,000 AI agents were working on research and engineering tasks. Oversight mechanisms are in place, and the firm has pledged to bring in external third‑party evaluators who will sit inside the company to audit safety practices.

The debate is far from settled. While some tech figures and even President Donald Trump push back against calls for restraint, Anthropic’s latest move shows a willingness to both lean into AI’s capabilities and shine a light on the process—hoping that transparency will give society a better chance to decide how—and how fast—we move forward.