Anthropic on Thursday outlined a proposal that frontier AI labs should prepare a coordinated, verifiable method to slow or temporarily pause development in the event that advanced systems begin to improve themselves at a rate society cannot safely handle. The company described the prospect of AI that can substantially build or iterate on itself as a major technological milestone that could also raise the risk of humans losing control over those systems.
In its statement, Anthropic warned that when systems are capable of producing their own successors, the approaches used to secure them, the ways they are monitored, and the techniques employed to shape their behavior become increasingly critical. To illustrate how autonomous development is already playing a role in its operations, the company said that, as of May, more than 80% of the code merged into its codebase had been authored by Claude.
Anthropic argued that 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, however, cautioned that isolated or poorly coordinated attempts to slow progress could be counterproductive if other actors continue advancing without similar constraints, potentially undermining overall safety.
To be meaningful, Anthropic said, a pause would need buy-in from "multiple well-resourced labs" operating at the technological frontier, and it would require agreed rules on what conditions should trigger a pause, what conditions would lift it, and who would oversee compliance. The company noted that a single firm's unilateral pause would be simpler to implement but would likely have limited effect beyond shifting leadership and would not substitute for broader global deliberation.
Anthropic also disclosed that its research arm, the Anthropic Institute, intends to study and help construct the technical and institutional systems that would be necessary to support a slowdown. In the months ahead, the company plans to convene discussions that bring together policymakers, researchers, civil society groups and other AI firms to examine key questions, including how to manage risks such as recursive self-improvement and how to strengthen coordination mechanisms.
Separately, Anthropic noted recent corporate developments: last month it closed a fundraising round that valued the company at $965 billion, and it confidentially filed for a U.S. initial public offering on Monday.
Context and implications
- The company emphasizes both technical research and multistakeholder governance as necessary components for any effective slowdown mechanism.
- Anthropic highlights the practical challenge that unilateral pauses pose: they are easier to enact but carry limited safety benefits if other advanced labs continue unfettered development.
- Its planned convenings aim to address coordination, triggers, oversight and the technical safeguards needed to make a pause credible and enforceable.