- Anthropic revealed AI may autonomously build itself and create more powerful successors
- Over 80% of Anthropic’s codebase by May 2026 was authored by its AI system Claude
- Recursive self-improvement raises oversight, security, and alignment challenges for AI
As if it couldn't get any more dystopian, Anthropic - the AI company behind Claude - just revealed that AI could start building itself and create its successor. The company also warned that "full recursive self-improvement also might increase the risks of humans losing control over AI systems." Simply put, recursive self-improvement means a stage where AI is able to improve itself without any human intervention and create a more powerful successor.
The company said that as it increasingly delegates AI developments to AI systems, there are signals that with enough compute it leads to a future where "an AI system (is) capable of fully autonomously designing and developing its own successor."
In fact, as of May 2026, more than 80% of the code merged into its
codebase was authored by Claude, the company revealed. A codebase serves as the single 'master plan' that developers use to build, test, and maintain the software
Anthropic, an outlier among AI companies, often warns of the ills of rapid artificial intelligence development. It said technical trends suggest AI systems are going to become much more capable in the coming years. "These trends have huge implications. AI that can build itself would be a major development in the history of technology-one that could bring enormous good for the world in science, healthcare, and beyond," it said.
However, it warned that "If systems are capable of fully building their own successors, the ways we secure them, monitor them, and shape their behavior all grow much more important."
Sagar Vishnoi, co-founder of Future Shift Labs also highlighted that as AI increasingly writes the code that powers future AI systems, the challenge moves from building capability to governing it. "Recursive self-improvement could significantly accelerate innovation, but it also raises important questions around oversight, security, and accountability. The real test for the industry will be ensuring that increasingly autonomous systems remain aligned with human priorities as they become more capable," he said
The Pause Clause
Anthropic argued that all AI companies need to come together to decide to "pause" frontier AI development, but warned that, "if a slowdown simply lets the least cautious actors catch up technologically, it could leave everyone less safe."
"We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology," it said.
This could perhaps be the hardest task of all, highlighted Dr Srinivas Padmanabhuni, CTO, AiEnsured. "Sounds good in theory but difficult to practice. Can we motivate all foundational model builders to pause? Very difficult to convince all, as everyone will feel the pressure of having been left behind," he said.
The AI giant's research arm, Anthropic Institute, plans to study and help build systems that would be necessary to support a slowdown. In the coming months, it plans to organise discussions involving policymakers, researchers, civil society groups and other AI firms to examine key questions around how to tackle the issue at hand. These questions include how to manage AI-related risks such as recursive self-improvement and how to improve mechanisms for coordination.
AI educator and founder of The Cutting Edge Group, Ansh Mehra lauded Anthropic's initiative and said tech companies should agree to a self-imposed global pause on LLM (large language model) releases for the next six months and focus a lot more on driving literacy, adoption and integration.
"We saw similar measures in the 1970s during the age of DNA technology, when scientists came together to first debate safety protocols instead of just rushing advancements.
"There has to be some sort of an LLM Treaty where nations agree to upskill a certain percentage of their population before releasing any new frontier models," Mehra pointed out.
However, just like Padmanabhuni, he remained skeptical about how these tech giants locked in the AI arms race would agree to this.
Pawan Prabhat, co-founder and President of ShortHills AI had a slightly different stance on fears of humans losing control to AI. "I am not worried that it will lead to a situation similar to The Matrix or Terminator because human beings can always pull off the plug if AI threatens its existence."
He however, warned of a trade off. "A self-improving omnipotent AI can do immense good to humanity but it will not come without its social costs."
