Anthropic Calls for AI Slowdown as Systems Begin Writing Their Own Code
Anthropic, maker of the Claude chatbot, says AI systems may soon reach a point where they can design and build their own successors with minimal human involvement. The company is calling on rival labs and governments to coordinate a slowdown in frontier AI development to give society time to catch up.
In a June 4 blog post, Anthropic said such recursive self-improvement could increase the risk of humans losing control of the technology. The company proposed what it called "a global coordination mechanism"-loosely modeled on arms-control agreements-to pause or slow development.
The warning centers on concrete changes within Anthropic itself. Claude now writes more than 80 percent of the code merged into the company's systems, up from low single digits before Claude Code launched in early 2025. Engineers are shipping roughly eight times as much code per quarter as they did a few years ago.
"We are not there yet, and recursive self-improvement is not inevitable," Anthropic said. "But it could come sooner than most institutions are prepared for."
The Coordination Problem
Any meaningful slowdown would require leading AI labs and multiple governments to accept identical limits simultaneously. No treaty exists to enforce such an agreement, and competition between companies is intensifying.
Anthropic offered few specifics on how verification would work or which institutions would need to participate. The company said it would spend coming months convening governments, researchers, and rival AI companies to explore whether a coordinated pause could function in practice.
Skeptics Question the Sincerity
Critics say the proposal rings hollow. Noah Giansiracusa, an associate professor of mathematics at Bentley University, said a slowdown is "literally impossible" and questioned whether Anthropic genuinely intends to brake its own progress.
The timing fuels suspicion. Anthropic unveiled a model called Mythos two months ago but declined to release it publicly, citing security concerns. The slowdown call came days after the company confidentially filed for an initial public offering and following a funding round that valued it at close to $1 trillion.
Mark Riedl, a professor at Georgia Tech's School of Interactive Computing, said the major AI companies are "jumping on the 'recursive self-improvement' hype train." Some researchers view such announcements as regulatory strategy-drawing scrutiny to the frontier while Anthropic continues advancing toward it.
Anthropic did not respond to questions about how a brake would work or whether it has overstated what its systems can do.
What the Evidence Shows
The fact that Claude writes most of the code in Anthropic's systems does suggest the technology is becoming more useful for internal work. Whether this constitutes a warning sign depends on interpretation.
Giansiracusa sees progress, not peril. "I see it continuing to progress. Maybe things will speed up; maybe it won't," he said. The increase in AI-written code reflects capability gains, not necessarily a threshold moment requiring global coordination.
For researchers working on generative AI and LLM systems, the question is practical: what would a slowdown actually require, and is it achievable? Research into AI governance and alignment remains crucial regardless of whether industry-wide coordination materializes.
Your membership also unlocks:
