Anthropic called for the creation of a coordinated mechanism that would allow leading AI developers and governments to collectively slow or temporarily pause frontier AI development, warning that the technology is advancing toward the ability to build and improve itself without human involvement.
The call came in a lengthy blog post authored by Anthropic co-founder Jack Clark and Marina Favaro, leader of the Anthropic Institute, the company's research arm. The post argued that AI systems are already meaningfully accelerating AI development and that the industry may be approaching a point where those systems can design their own successors — a phenomenon the authors call "full recursive self-improvement."
"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," the company said.
Appearing on CNN Thursday night, Clark described the current moment as a car with only a gas pedal. "When I look down at the car we're driving, all I have is a gas pedal. I don't have a brake pedal, and surely at some point in the future we might want that option," he told CNN.
Anthropic pointed to its own internal data to illustrate how rapidly AI is reshaping the development process. Claude now accounts for more than 80% of code contributions to its codebase, Anthropic said, and engineers were absorbing eight times as much merged code daily in the second quarter of 2026 compared with two years earlier. A March 2026 internal survey of 130 employees found that the median respondent estimated producing roughly four times as much output with AI assistance as without it.
The company stopped short of calling for an immediate unilateral pause, arguing that such a step by a single lab would change who leads the field without creating the broader deliberative process the moment requires. Any credible slowdown, Anthropic said, would require multiple well-resourced labs at or near the frontier — across multiple countries — agreeing to stop under the same conditions, with each able to verify that the others have actually done so.
"Training runs are far easier to conceal than missile silos, their inputs are general-purpose, and the incentive to defect quietly is enormous, because whoever continues while others pause could inherit the lead," the company said. Clark invoked the history of nuclear diplomacy, arguing that countries locked in fierce rivalry had nonetheless managed to place limits on weapons technology — a precedent he said the AI industry should study.
Anthropic said it would organize conversations in the coming months with policymakers, researchers, civil society, and other AI companies to address questions raised by the post, and would publish the results.
The blog post arrives at a consequential moment for Anthropic. The company raised $65 billion in new funding at a $965 billion valuation in late May, and submitted a confidential IPO prospectus to the Securities and Exchange Commission earlier this week. A separate Bloomberg report noted that Anthropic had walked back an earlier internal pledge to halt development if conditions became dangerous, a decision the company framed as a response to a regulatory climate that had come to emphasize national competitiveness over precautionary measures.