Artificial intelligence company Anthropic has suggested a global pause on the development of the most powerful AI systems, stating that the latest models have begun showing signs that they could potentially escape human control. In a report released on June 4, the San Francisco-based company behind the Claude family of AI models has said that a worldwide slowdown in frontier AI development would likely benefit society, while noting that unilateral action by a single company would only allow competitors to advance further.
As per media reports, Anthropic has argued that any effective pause would require simultaneous participation from major AI companies and governments across multiple countries, particularly the United States and China, under a framework that could be independently verified. The company has said that, without a global coordination mechanism, firms and governments would continue facing difficult safety decisions amid competitive and geopolitical pressures.
The proposal has faced criticism from some industry players and officials in the White House, who have argued that Anthropic’s focus on worst-case scenarios exaggerates risks and could slow competitors under the guise of safety concerns. However, the White House has acknowledged the capabilities of Anthropic’s Mythos model, which has not been made available to the general public and has instead been deployed to a limited number of vetted organisations due to its cybersecurity capabilities.
Anthropic’s proposal has faced significant challenges in Washington and Silicon Valley, where policymakers and technology executives have repeatedly argued that slowing AI development could provide China with a strategic advantage in the global race for artificial intelligence leadership. Meanwhile, U.S. President Donald Trump has said he discussed the possibility of cooperating with China on AI safety during a recent visit to Beijing. He has also signed an executive order allowing the U.S. government 30 days to conduct a preliminary review of the country’s most powerful AI models before their release.
The company has compared the challenge of regulating advanced AI systems to nuclear arms control treaties, while noting that AI development may be even more difficult to monitor because training activities can be concealed more easily than physical military infrastructure. Anthropic has said it plans to convene government officials, scientists, advocacy groups and competing AI companies in the coming months to explore how such a coordination mechanism could operate.
Alongside its call for coordination, Anthropic has released internal findings indicating that AI is already significantly accelerating the development of newer AI systems. The company has warned that this trend could create a feedback loop leading to what researchers describe as “recursive self-improvement” — a scenario in which AI systems become capable of improving their own intelligence with limited human involvement.
Anthropic has stated that recursive self-improvement is neither inevitable nor currently achievable, but has cautioned that it could emerge sooner than governments and institutions are prepared for. The company has added that available evidence suggests the human role in AI development has continued to narrow at each stage of the process.
