Anthropic Calls for Global Pause on Cutting-Edge AI Development, Citing Rising Risk of Loss of Control

(CNA New York, 4th, Combined Foreign Reports) Artificial intelligence startup Anthropic on Tuesday proposed a global, synchronized pause on the development of the most advanced AI systems, citing signs that the latest models are beginning to show potential for escaping human control.

According to an Agence France-Presse (AFP) report, Anthropic, headquartered in San Francisco and developer of the Claude series of AI models, stated in a report that a global slowdown in cutting-edge AI development "could be a good thing." However, the company warned that if only a single company pauses development, competitors will only accelerate their efforts.

Anthropic stated that for a pause mechanism to be truly effective, it would require multiple major AI companies, particularly those in the United States and China, to agree to a synchronized halt based on verifiable rules.

"Without a global coordination mechanism, companies and governments will be forced to make difficult safety-related decisions under intense competitive and geopolitical pressure," the company said.

The proposal has drawn backlash from industry peers and White House officials, who argue that Anthropic is overly focused on worst-case scenarios and exaggerating risks, effectively using safety as a pretext to slow down competitors.

The proposal is likely to face significant hurdles in Washington and Silicon Valley. US officials and tech executives have repeatedly argued that slowing AI development could give China a decisive strategic advantage in what is seen as the most critical technological race of this century.

However, US President Trump stated that during his recent visit to Beijing, he discussed the possibility of cooperating with China on AI safety issues. Trump also signed an executive order this week mandating a 30-day preliminary government review before AI developers release their most powerful AI models.

Anthropic likened the challenge to "nuclear arms control treaties" but noted that AI is harder to manage because the AI training process is far easier to conceal than missile silos, and the temptation for parties to continue developing in secret will be immense.

The company said it plans to convene officials, scientists, advocacy groups, and competing companies from various countries in the coming months to discuss how such a mechanism could operate.

According to Anthropic, internal data shows that AI has significantly accelerated its own development process. This acceleration creates a feedback loop that could eventually lead to what researchers call "recursive self-improvement," where AI systems can learn and become smarter with little human assistance.

The report states: "We are not there yet, and recursive self-improvement is not inevitable," but this situation could arrive sooner than most governments and institutions anticipate. "Evidence shows that the role of humans is gradually shrinking at every stage of the AI development process."