Anthropic's Claude AI now leads 26% of the company's model research and development, completing most tasks "end-to-end from a high-level prompt" under human supervision. The disclosure, made this week, offers the clearest public benchmark yet for how close top labs are to recursive self-improvement - AI systems that build smarter versions of themselves.
The milestone arrives as a group of AI executives, including Anthropic CEO Dario Amodei, last weekend called for a coordinated slowdown on development pace. Their concern: models that learn to design their own successors could slip beyond human control faster than safety measures can keep up.
What recursive self-improvement actually means
Definitions vary across labs. Some call it RSI as soon as AI provides any feedback on model improvement. Others reserve the term for fully autonomous cycles where one model designs the next without human involvement.
Anthony Aguirre, president of the Future of Life Institute and a physics professor at UC Santa Cruz, described autonomous RSI as "AI that can improve itself designing the next version of the system, then the next version, and so on." He added, "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."
John Thickstun, an assistant professor of computer science at Cornell University, 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," he said. Past efforts by researchers like OpenAI co-founder Andrej Karpathy produced minor gains but no large creative leaps. That may be changing.
Aguirre pointed to Anthropic's trend line as evidence. "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."
Where the major labs stand
Anthropic has not said exactly how close it is to fully autonomous RSI. The company has committed to slowing or pausing development if competitors do the same in a verifiable manner.
OpenAI announced this month that it built an automated "research intern" capable of completing tasks that would take a skilled researcher several days. The company is targeting a fully automated AI researcher by March 2028 but cautioned that rapid RSI is not necessarily an outcome to 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," OpenAI said in a blog post.
Elon Musk told investors in March that 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." He projected full automation by the end of 2025 and "not later" than 2027.
Microsoft AI CEO Mustafa Suleyman has charted a different course, describing a vision of "humanist superintelligence" - advanced capabilities that remain "carefully calibrated, contextualized, within limits" rather than an unbounded autonomous entity.
The safety question no lab has answered
OpenAI said this month it does not yet know how to "safely get all the way to aligned, full RSI," noting that more capable systems become harder to monitor. Still, the company considers pursuing RSI valuable because "an automated AI researcher can also be an automated safety or alignment researcher."
Thickstun sees the fear of runaway superintelligence as the core anxiety driving RSI debates. The practical reality, he argued, is that recursive improvement has been underway for years through human-AI collaboration. What's shifting now is the ratio of human to machine involvement - and the speed at which that ratio is tilting.
Professionals in AI R&D Engineering Courses are already working with the generation of tools that feed into these self-improving loops. Understanding how models contribute to their own design is becoming a practical skill, not a theoretical concern.
Why this matters for executives and strategy leaders
The move toward RSI changes the calculus for any organization building on top of frontier models. A system that designs its successor can compress development timelines in ways that outpace governance, compliance, and safety review processes. The labs themselves are publicly split on how fast to go and what guardrails to install. For leaders in regulated industries - healthcare, insurance, product development - the gap between model capability and verified safety is the metric to watch, not just the capability itself.
Your membership also unlocks:
