AI systems are not yet building their own successors. Anthropic says the industry should start preparing as if they might.
In a new report, the company warned that AI is taking on an increasing share of coding, debugging, and research work within frontier labs, including Anthropic itself. If that trend continues, future systems could become capable of helping design and develop more powerful models faster than institutions can test, regulate, or contain them.
The concern is not a movie-style robot rebellion. It is a governance problem: What happens when the tools used to build AI begin accelerating the next generation of AI?
Claude is already writing much of Anthropic’s code
Anthropic said in a new report from Anthropic Institute that humans still drive model development. Still, AI systems now handle a growing share of coding, debugging, and technical research inside the company.
The company stated that Anthropic engineers now ship, on average, eight times as much code per quarter as they did from 2021 to 2025. As of May 2026, Claude authored over 80% of the code merged into its codebase, according to the report.
The trend points toward a possible future in which an AI system can autonomously design, build, and train a more capable successor. Anthropic called this “recursive self-improvement,” while cautioning that the industry is not there yet and may never arrive there.
Such developments could make oversight harder. If AI systems help build their own successors, labs and regulators may have less time to test whether those systems are secure, accurate, and aligned with human goals.
Oversight becomes the central question
Reuters reported that Anthropic wants frontier AI developers to create a coordinated, verifiable way to slow or pause development if advanced systems begin improving themselves faster than society can manage the risks.
Anthropic cautioned that a pause by a single company would have a limited impact if other well-funded labs continued to race ahead. The company noted that a useful slowdown would require coordination among multiple frontier labs, clear rules for what triggers or ends a pause, and trusted oversight.
According to Axios, Anthropic plans to brief lawmakers on recursive self-improvement in the coming months. Jack Clark, Anthropic’s co-founder and head of policy, told Axios, “We’ve always found that the best thing to do is to socialize the concept and basically give people a sense of what’s coming.”
From sci-fi warning to enterprise risk
Interesting Engineering likened the scenario to “Age of Ultron,” a pop-culture comparison that fits the anxiety around machines helping create more powerful machines. The real-world issue, however, is less theatrical — AI companies are increasingly using AI systems to assist with coding, research, and other development work.
Anthropic said current systems still fall short of full autonomy. Humans choose research goals, judge results, and decide which systems move forward. But the company said AI is taking on a growing share of development work, a trend it believes could eventually lead to systems capable of helping build more capable successors.
Anthropic plans to discuss the issue with policymakers, researchers, civil society groups, and other AI companies in the coming months as it explores what safeguards may be needed if AI systems become more involved in developing future generations of models.
Read more: Anthropic Files to Go Public, Setting Up Blockbuster AI IPO