The prospect of artificial intelligence models teaching themselves to become more efficient and capable — once distant ambition among technology researchers — is drawing closer to reality.

As the technology advances, developers say it is approaching “recursive self-improvement,” or RSI, a stage in which AI models find ways to improve themselves and build their successors. Technology executives say the shift could bring advances in science and medicine, but it also carries risks.

The uncertainty over where it all could lead sits at the heart of growing fears about AI evading human control, and of possible threats to humanity. Those concerns led several AI industry figures to join a call last weekend to slow the technology's pace of growth.

Anthropic this week detailed how its model Claude is helping the company develop the next, more intelligent version of itself. Claude now leads 26% of Anthropic's model research and development, which the company said means it can complete most of a given task “end-to-end from a high-level prompt” while still operating under human supervision. The models are not working completely autonomously — at least not yet.

Competing Definitions of RSI

Leading AI companies define recursive self-improvement differently. Some define it as any feedback from AI on model improvement, while others define it as AI working toward that goal fully autonomously. The distinction matters: because there is no shared threshold for what counts as RSI, progress claims from different labs are not all measuring the same thing.

Autonomous recursive self-improvement essentially means AI that can improve itself by designing the next version of the system, then the next version, and so on, according to Anthony Aguirre, president and CEO of the nonprofit Future of Life Institute and a physics professor at the University of California, Santa Cruz.

“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.

Experts Weigh Runaway Risk Against a Gradual Reality

The fear surrounding RSI is based largely on the possibility of a runaway superintelligence emerging from the process — a scenario in which self-improving systems advance beyond humans' ability to monitor or control them — said John Thickstun, an assistant professor of computer science at Cornell University who studies methods for controlling the behavior of AI models. But he said a more grounded view suggests a form of recursive self-improvement has been underway in AI development for some time.

“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.

Prominent AI researchers — including OpenAI co-founder Andrej Karpathy — have for years experimented with getting AI models to train and improve new AI systems. Those efforts have produced minor improvements but no big creative leaps, Thickstun said.

Aguirre, however, said AI companies today are much closer to achieving those larger leaps in improvement.

“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 said. “And 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.”

Anthropic's announcement gave the public — and other labs — some insight into RSI progress, and the company encouraged its competitors to share similar metrics. Figures that competitors publish in response could give the industry a common baseline for comparing progress toward autonomous research. Still, Anthropic has not expressly said how close it is to achieving fully autonomous model improvement.

OpenAI's “Research Intern” and a 2028 Goal

ChatGPT maker OpenAI announced this month that it has developed an automated “research intern,” which it defines as a system capable of carrying out well-defined research tasks under human direction, including “tasks that would take a skilled researcher a few days.” The company has said it is moving forward with the goal of creating an automated AI “researcher” by March 2028. The date gives observers a concrete, company-stated milestone against which to measure that progress.

In that announcement, OpenAI said that while RSI can help align models' actions with human values and intentions, that does not mean “rapid RSI is necessarily an outcome we should pursue.”

“Whether and how to proceed must depend on our ability to preserve human control and on informed democratic choices about the benefits and risks,” the company said in a blog post.

Musk Signals a Faster Timeline at xAI

Elon Musk appears more eager to forge ahead. Speaking in March about xAI's Grok models, he said “humans are gradually getting less and less in the loop” on model improvement and that “every successive model is built by the one before it,” while clarifying that the process was not yet fully automated. That target might be reached by the end of this year, he added, “but not later” than 2027 — a timeline that runs ahead of OpenAI's March 2028 goal, though the two companies describe the milestone differently.

Microsoft's “Humanist” Alternative

Microsoft and some other leading AI companies appear to be taking a different approach. Mustafa Suleyman, CEO of Microsoft AI, has said the company is moving toward “humanist superintelligence” — advanced AI capabilities placed in service of people and humanity at large. In a 2025 essay, Suleyman said this would not mean “an unbounded and unlimited entity with high degrees of autonomy,” but rather AI that is “carefully calibrated, contextualized, within limits.”

Safety Measures Struggle to Keep Pace

A key challenge facing labs — one they have confronted essentially since the technology's inception — is ensuring their safety measures advance alongside their models' capabilities.

Divisions have emerged across the tech industry over calls for a coordinated AI slowdown for safety, and not every major player in the AI space has specifically commented on its path forward with RSI.

Anthropic, which has been a leading voice in calls for pacing, has said it would slow or temporarily pause its development work — assuming its global competitors did so as well, and in a “verifiable manner.”

OpenAI explicitly said this month that it does not yet know how to “safely get all the way to aligned, full RSI,” adding that the company “cannot assume that progress in alignment and safety will keep pace.” More capable systems can become harder to monitor, it continued, but pursuing RSI remains a goal the company values because an “automated AI researcher can also be an automated safety or alignment researcher.”

This story was originally featured on Fortune.com.