Anthropic is calling for top artificial intelligence labs to weigh slowing the pace of development, suggesting that AI systems are advancing so rapidly that they may soon be able to improve themselves without human intervention in ways that could pose significant societal risks.

The ability to slow global AI development would “likely be a good thing,” the company said Thursday in a blog post that disclosed internal data documenting how quickly its most advanced models are improving.

The post, written by the head of its internal research institute and head of policy, noted that model advances appear to be on a path toward “recursive self-improvement,” when AI systems can improve on their own without human intervention. Some AI insiders have seen that threshold as a potential marker of danger and enormous societal upheaval.

“We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology,” the post, written by Marina Favaro and Jack Clark , says. It proposes a global agreement on how to potentially slow development and a mechanism for verifying that competitors are respecting it.

The post cautions that recursive self-improvement hasn’t yet happened and isn’t inevitable, “but could come sooner than most institutions are prepared for.”

Anthropic recently concluded a fundraising round that valued the company at almost $1 trillion and filed confidential paperwork to begin the process of publicly listing its shares . The company has recently emerged as the front-runner in a ferocious competition for AI supremacy with ChatGPT-maker OpenAI, which is also expected to file paperwork for an initial public offering soon.

Anthropic’s run rate, a figure commonly used by startups that forecasts annual revenue based on short-term sales, is on track to reach $50 billion in annualized revenue by the end of this month, up from $9 billion at the end of 2025.

The company, which has emphasized AI safety from its founding, has long faced criticism that its policy work is designed to slow the AI advances of competitors. David Sacks , a venture capital investor and informal adviser to President Trump , has accused Anthropic’s leaders of running a “regulatory capture agenda.”

On a recent podcast, Sacks said the “reg capture agenda” in Washington could lead to an effort to ban so-called open-source models, essentially versions of AI systems that are far cheaper for organizations to use and develop internally.

Others have suggested that Anthropic’s warnings about the dangerous potential of its own tools could also be considered a marketing ploy. Such skeptics point to Anthropic’s decision to limit its release of a powerful “Mythos” cybersecurity model capable of finding bugs and problems as a handy way to tout the capabilities of its products.

Anthropic’s leaders have said they take safety seriously and are working to create more discussion of risks.

Ethan Mollick, a professor at the University of Pennsylvania’s Wharton School and an influential scholar on AI transformation, suggested on X that people read the new Anthropic post.

“There is a bit of navel-gazing, some marketing, and a lot of very sincere beliefs about what Anthropic thinks is likely in the near future of AI that you probably want to be aware of,” he wrote about Anthropic’s post.

The AI industry has been divided for some time on how close current models are to reaching benchmarks like “artificial general intelligence,” or AGI, a level of intelligence that is comparable to humans, or recursive self-improvement.

Some scholars, such as Yann LeCun , former chief AI scientist at Meta Platforms and an AI pioneer, has argued that frontier systems based on large-language models won’t ever be capable of making the leap to rivaling human intelligence. While seeing AI models as powerful tools, he has compared them to the intelligence of a cat and sparred with researchers who fear that AI poses an existential risk to humanity.

Anthropic’s leaders, including Chief Executive Dario Amodei , have warned about the potential for dangerous impacts from AI for years and sought widespread societal collaboration to address risks. Amodei has warned that AI could worsen inequality and eliminate as many as half of entry-level white-collar jobs .

Amodei has also warned that it is plausible that powerful AI systems, which he expects to exist in the near future, could develop destructive tendencies in unpredictable ways. In an essay on his own website in January, Amodei suggested that training of AI systems with science-fiction narratives of AI rebellion could, for instance, end up causing real AIs to rebel.

Clark, the blog post’s co-author and an Anthropic co-founder, has frequently spoken about potential AI risks and the potential impact of AI systems reaching the point of “recursive self-improvement,” which he has said would transform the world.

“That class of technology has never existed before, and yet I believe this could happen within the next two years, and possibly sooner,” he said at a lecture last month in London, according to remarks he shared in a newsletter.

“In the absence of a coordinated, global slowdown, we are left with the current situation: powerful technology being developed at breakneck speed by a variety of actors in a variety of countries, locked in a competition with one another where commercial and geopolitical rivalries are drowning out the larger existential-to-the-species aspects of the technology being built,” he said.

Thursday’s blog post says the Anthropic Institute, an in-house research shop dedicated to understanding and shaping how AI will impact the world, will conduct research in collaboration with others to “help build the systems that a credible slowdown or pause would require.”

The post said that a pause or slowdown would only make sense if it were widely respected, which would require a verification regime to make sure everyone has stopped. The company likened the problem to nuclear-weapons treaties—but acknowledged stopping cheating would be even thornier.

“Training runs are far easier to conceal than missile silos,” the blog post read, adding, “whoever continues while others pause could inherit the lead.”

Anthropic plans to organize conversations in the coming months with policymakers, researchers and others to help answer questions around recursive self-improvement and a verification system: “The window to investigate the questions together is here, and people outside AI companies should be involved in this deliberation,” the post said.

Write to Bradley Olson at bradley.olson@wsj.com and Sam Schechner at Sam.Schechner@wsj.com