As AI helps build its successors, Anthropic calls for a way to slow development

Anthropic has called for the creation of an international mechanism that would allow leading AI developers to slow or temporarily halt the development of frontier artificial intelligence systems if the technology begins advancing faster than society can safely absorb.

The proposal, outlined in a paper published by the Anthropic Institute, comes as the company argues that AI systems are increasingly contributing to the development of their successors, a process researchers refer to as recursive self-improvement. The company said the ability to pause development, should it become necessary, may prove difficult to achieve in practice because of intense commercial and geopolitical competition among AI developers.

“If it were possible to effectively slow the development of this technology to give ourselves more time to deal with its immense implications, we think that would likely be a good thing,” Anthropic wrote. But it warned that a slowdown by only a handful of companies could backfire if “the least cautious actors catch up technologically,” potentially leaving the broader ecosystem less safe.

Claude can improve itself

Anthropic noted that its engineers on average ship 8x as much code per quarter as they did from 2021-2025. It said that an AI system is capable of fully autonomously designing and developing its own successor, called recursive self-improvement. “We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for.”

 “The technical trends discussed in this piece suggest that AI systems are going to become much more capable in coming years. These trends have huge implications,” wrote Marina Favaro, Lead at The Anthropic Institute and Jack Clark, Co-founder of Anthropic.

“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. But full recursive self-improvement also might increase the risks of humans losing control over AI systems. If systems are capable of fully building their own successors, the ways we secure them, monitor them, and shape their behaviour all grow much more important,” the report read.

Rather than advocating a unilateral halt, Anthropic argues that any meaningful pause would need to involve multiple frontier AI laboratories operating across different countries. Such an arrangement would also require mechanisms allowing participants to verify that rivals had genuinely stopped developing advanced systems rather than continuing in secret.

“A meaningful slowdown or pause would require multiple well-resourced labs at or near the frontier, in multiple countries, agreeing to stop under the same conditions,” the company said.

 Advance AI development harder to monitor

The challenge, according to Anthropic, is that advanced AI development is considerably harder to monitor than many other strategically sensitive technologies. Unlike missile silos or nuclear facilities, large AI training runs can be concealed more easily, rely on widely available computing infrastructure and create powerful incentives for companies or governments to quietly continue development while competitors pause

“Whoever continues while others pause could inherit the lead,” the report noted. Anthropic acknowledged that creating a globally enforceable framework would be extraordinarily difficult. The company pointed to past arms-control agreements as evidence that verification systems can be built, but noted that such arrangements often took decades to establish.

“Those regimes took decades to build both the infrastructure and the trust. We don’t have that long,” it wrote. The proposal follows Anthropic’s broader warning that AI could eventually reach a stage where it plays a substantial role in designing and developing future generations of AI systems. While the company stressed that fully autonomous recursive self-improvement remains speculative, it argued that governments, researchers, civil society groups and AI companies should begin preparing for the possibility now rather than after the technology arrives.

To that end, Anthropic said it plans to convene discussions involving policymakers, researchers, civil society organisations and competing AI developers to explore how coordination mechanisms might work and under what conditions a slowdown could be triggered.

“The window to investigate the questions together is here,” the company wrote, adding that people outside AI companies should play a role in determining how society responds to increasingly capable AI systems.