Anthropic's Claude now helps design its own smarter replacement

Anthropic’s Claude has taken on a new kind of job: helping design the next, smarter model that could one day replace it.

This week, the company disclosed that Claude now leads 26% of Anthropic’s model research and development.

That means Claude can complete most of a given task “end-to-end from a high-level prompt,” although a human still supervises the work.

The models are not working on their own — not yet — but the shift offers a rare look at how close the field is to what researchers call recursive self-improvement.

What Is Recursive Self-Improvement?

Recursive self-improvement, or RSI, is the idea that an AI system can improve itself and then help build its successor. For years, it was seen as a distant ambition.

Today, leading labs are closer, though they define the term differently.

Some count any feedback from AI on model improvement as RSI; others reserve it for AI working toward that goal fully autonomously.

Anthony Aguirre, president and CEO of the Future of Life Institute and a physics professor at the University of California, Santa Cruz, warns about the speed of that process. “The really important thing here is that as AI is doing more of it, it gets faster, because AI operates just much, much more quickly than the humans do,” he said.

John Thickstun, an assistant professor of computer science at Cornell University who studies ways to control AI behavior, said the fear around RSI is based largely on a runaway superintelligence emerging from that process.

But he offered a more grounded view. “We have already, for years, been using these models in supportive roles for creating the next version of these models.

So people use the past generation of models to write code for the AI systems that then create the next generation,” he said.

Earlier efforts, including experiments by people like OpenAI co-founder Andrej Karpathy, led to minor improvements but not major creative leaps, Thickstun said.

Aguirre thinks Anthropic’s data shows that may be changing: “You can see in these plots from Anthropic over time, more and more of research is being done by the AI and it’s becoming closer and closer to fully autonomous.” He added: “The result of that success, ultimately is something that is, I think, extremely scary.

I think this is probably the worst idea in the history of humanity to do this. And yes, they’re doing it.”

Other Labs Are Moving Too

Anthropic has not said how close it is to fully autonomous model improvement. It did encourage competitors to share similar numbers.

OpenAI this month announced an automated “research intern,” which it defines as a system that can carry out well-defined research tasks under human direction, including “tasks that would take a skilled researcher a few days.” OpenAI says it is moving toward an automated AI “researcher” by March 2028.

OpenAI also offered a cautious note.

It said it does not yet know how to “safely get all the way to aligned, full RSI,” and it “cannot assume that progress in alignment and safety will keep pace.” The company said that while RSI can help align models with human values, that doesn’t mean “rapid RSI is necessarily an outcome we should pursue.” It added that “whether and how to proceed must depend on our ability to preserve human control and on informed democratic choices about the benefits and risks.” Still, OpenAI said it values pursuing RSI because “an automated AI researcher can also be an automated safety or alignment researcher.”

Elon Musk has signaled a faster timeline.

In March, he said for xAI’s Grok models, “humans are gradually getting less and less in the loop” and “every successive model is built by the one before it,” though he clarified the process is not yet fully automated.

He said that target could be reached by the end of this year, “but not later” than 2027.

Microsoft appears to be taking a different path.

Mustafa Suleyman, CEO of Microsoft AI, has described a “humanist superintelligence”: advanced AI in service of people, but not “an unbounded and unlimited entity with high degrees of autonomy.” Instead, he said, AI should be “carefully calibrated, contextualized, within limits.”

What This Means Beyond the Lab

The public disagreement over RSI is not merely technical.

If AI is writing more of the code and making more design choices behind the next generation of models, the pace of progress could speed up in ways that are hard to predict.

That could bring advances in science and medicine, as some tech executives argue.

It also increases the stakes of safety, because more capable systems may become harder to monitor, and labs are not moving at the same speed or with the same guardrails.

One way to read Anthropic’s disclosure is as a transparency milestone: a leading lab gave the public a measurable, though partial, look at how much AI is now involved in building itself.

For everyone else, that means the debate about slowing down AI development is no longer an abstract thought experiment.

Several prominent AI figures recently joined a call to slow the technology’s growth, but actual practice is already pushing the other direction in some labs.

For ordinary people and small businesses, the practical advice is straightforward. Do not assume today’s AI tools work the way they did six months ago.

Keep human oversight in important workflows, especially when a task affects money, health, or decisions about people.

Ask vendors whether their products are built with human review and whether any part of the model was itself designed by another AI.

Demand clear answers on safety practices before handing over sensitive responsibilities.

Anthropic has said it would slow or pause its own development work if global competitors did the same, and did so in a “verifiable manner.” That kind of collective restraint has not happened yet.

The next few quarters may reveal whether RSI becomes a carefully managed tool or a race with fewer checks.

Source: Kaitlyn Huamani and Barbara Ortutay, Associated Press