Anthropic Proposes Global AI Development Pause Amid Control Concerns

Anthropic, the San Francisco company behind the Claude AI models, has issued a striking proposal: a coordinated global slowdown in developing the most powerful artificial intelligence systems. The company warned Thursday that current models are beginning to exhibit signs they could surpass human control.

The recommendation, detailed in a new company report first covered by France 24, acknowledges the proposal faces significant practical and political obstacles. Anthropic emphasized that a pause would only work if major AI companies across multiple countries—particularly the United States and China—agreed to simultaneous restrictions with verifiable enforcement mechanisms.

The coordination challenge

Anthropod compared the coordination problem to nuclear arms control treaties, but noted AI development presents even steeper verification challenges. Unlike missile silos, AI training infrastructure is relatively easy to conceal, and competitive pressures create strong incentives for covert development.

"Without a global coordination mechanism, companies and governments will have to make difficult decisions about safety while under competitive and geopolitical pressures," the company stated in its report.

The proposal would allow societal structures and alignment research—the field focused on ensuring AI systems behave as intended—to catch up with rapid technological advances.

Industry and government tensions

The call has drawn criticism from competitors and White House officials who argue Anthropic overstates risks and uses safety concerns to slow rivals. The proposal runs counter to prevailing sentiment in Washington and Silicon Valley, where leaders frame AI development as a strategic competition with China that cannot afford delays.

Yet the White House has acknowledged the capabilities of Anthropic's Mythos model, which remains restricted to vetted organizations due to its cybersecurity capabilities rather than being released publicly.

President Donald Trump signed an executive order this week requiring a 30-day government review of the most powerful US AI models before release. Trump also indicated he discussed potential AI safety cooperation with China during his recent Beijing visit.

The recursive improvement concern

Anthropod's report includes internal data showing AI systems are already accelerating their own development cycles, creating what the company describes as a narrowing human role at each stage of the AI development process.

This acceleration raises concerns about "recursive self-improvement"—scenarios where AI systems become capable of enhancing their own capabilities with minimal human involvement. While Anthropic stated this threshold has not been reached and is not inevitable, the company warned it could arrive before governments and institutions are prepared.

Why it matters

This marks one of the most explicit calls from a leading AI company for industry-wide development constraints. Unlike previous safety discussions focused on voluntary commitments or individual company policies, Anthropic is proposing enforceable international coordination—a framework that would require unprecedented cooperation between geopolitical rivals and competitive businesses. The proposal tests whether the AI industry can implement meaningful safety measures when they conflict with commercial and strategic incentives.

Anthropod plans to convene government officials, scientists, advocacy groups, and competing AI firms in coming months to explore implementation pathways.

Details were first reported by France 24 with AFP.