Anthropic calls for global coordination to pause frontier AI

Anthropic on Thursday called for international coordination to enable a potential pause in frontier artificial intelligence development, stating that society may struggle to keep pace with rapid technological advances. The California-based company said it would support such a measure only if multiple well-resourced laboratories across several countries agreed to halt work simultaneously under verifiable conditions, according to the report.

Verifiable multilateral halt

The company emphasized that any meaningful pause would require leading developers at or near the technological frontier to stop under identical conditions while maintaining robust mechanisms to verify genuine compliance. "If such systems existed, we expect that we would slow down or temporarily pause, if other developers at or near the frontier also did so in a verifiable manner," Anthropic said.

Erosion of human oversight

Anthropic noted that AI systems are increasingly assuming tasks previously performed by human engineers, with evidence suggesting that "the human role is narrowing at each step in the AI development process." More than 80 percent of code merged into the company's production codebase is now written by Claude, its AI assistant, while engineers produce eight times more code per quarter than during the 2021-2025 period, the report stated. "Put simply: the doing now costs almost nothing in human time," the company added, predicting that humans will eventually stop writing code entirely and shift exclusively to review functions.

Recursive improvement risks

The company warned that these trends could lead to "recursive self-improvement," where AI systems design and develop their own successors without human intervention. While such capabilities could accelerate scientific advances, Anthropic cautioned that full recursive self-improvement might increase the risks of humans losing control over AI systems. "The window to investigate the questions together is here," Anthropic said, urging policymakers, researchers, and civil society groups to take part in discussions over AI governance.