Artificial intelligence firm Anthropic has called for the possibility of a global pause in the development of the most advanced AI systems, warning that increasingly powerful models are beginning to display capabilities that could eventually become difficult for humans to control.
The San Francisco-based company, which develops the Claude family of AI models, said in a new report that a worldwide slowdown in frontier AI development would likely benefit society by allowing safety research and governance structures to keep pace with technological progress, reports RTE.
However, Anthropic argued that such a pause would only be effective if it were implemented collectively by major AI developers and governments around the world.
The company said that if a single organisation chose to slow development on its own, competitors would simply continue advancing their systems and gain a strategic advantage, reports RTE.
According to Anthropic, any meaningful pause would require cooperation between leading AI companies and governments, particularly in the United States and China, alongside mechanisms to verify compliance.
The company warned that without international coordination, both governments and technology firms will continue facing difficult decisions about safety while under pressure from commercial competition and geopolitical rivalries, reports RTE.
Anthropic’s position has attracted criticism from some figures within the technology sector and from officials in the White House, who argue that the company places too much emphasis on worst-case scenarios and may be using safety concerns as a way to slow competitors.
Despite those criticisms, the White House has acknowledged the capabilities of Anthropic’s Mythos model, which has not been released to the public because of its advanced cybersecurity functions and is currently available only to a limited number of approved organisations, reports RTE.
The proposal is expected to face resistance in both Washington and Silicon Valley, where policymakers and industry leaders have frequently argued that slowing AI development could allow China to gain a significant advantage in what many view as one of the century’s most important technological races.
US President Donald Trump recently said he discussed potential cooperation with China on AI safety during a visit to Beijing and also signed an executive order requiring a preliminary government review of the most powerful American AI models before they are released, reports RTE.
Anthropic compared the challenge of regulating advanced AI to international nuclear arms-control agreements but noted that AI development may be even harder to monitor because training systems can be concealed far more easily than physical military infrastructure.
The company said it intends to bring together government representatives, scientists, advocacy groups and competing AI firms in the coming months to explore how a coordinated international framework might operate, reports RTE.
Anthropic also pointed to internal research suggesting that AI systems are increasingly helping to accelerate the development of newer AI systems, creating what it described as a potentially powerful feedback loop.
The report warned that this trend could eventually lead to so-called “recursive self-improvement”, where AI systems become capable of substantially improving their own intelligence with limited human involvement, reports RTE.
While Anthropic stressed that such a scenario has not yet emerged and may never fully materialise, it cautioned that progress toward that possibility could occur faster than governments and institutions are currently prepared for.
The company concluded that available evidence suggests human involvement is gradually shrinking at each stage of the AI development process, highlighting the need for greater attention to long-term governance and safety measures, reports RTE.
Tell us your thoughts in the Facebook post and share this with your friends.
