Anthropic says AI is advancing so fast that major research institutions may need to slow down.
Researchers at the company’s research arm, Anthropic Institute, said in a blog post published Thursday that the AI is already accelerating the development of new AI models very rapidly and could eventually help build its own successor models.
“We believe the world is better off with the option of slowing or temporarily halting the development of frontier AI so that social structure and coordination research can catch up with technological advances,” the company said in a statement.
Anthropic’s warning comes as AI is reshaping the technology industry and the entire economy at breakneck speed.
In recent months, executives have touted AI’s increasing ability to perform tasks once performed by humans, from writing software code to handling hiring and customer service tasks. Google says AI currently generates 75% of its code, but Mercor CEO Brendan Foody recently said his startup spends more on AI tokens than on employee salaries.
At the same time, more companies are linking layoffs and restructuring to AI-enabled efficiency gains.
Anthropic pointed to its own internal data to show how much work is already being delegated to AI.
More than 80% of the code merged into the codebase is now written by Claude, and the typical engineer was merging eight times more code per day in the second quarter of 2026 than in 2024.
One of the employees quoted in the blog post said that there are days when everything goes so well that “you can’t help but think that nothing you do matters,” while another said it’s been about five months since he last wrote any code himself.
The report also notes that AI systems are increasingly capable of handling engineering and research tasks that once required humans.
The company said that while current models still struggle to make higher-level decisions and decide which problems are worth solving, AI capabilities are rapidly advancing.
Anthropic stopped short of calling for an immediate suspension.
But he said any significant slowdown would require coordination between frontier AI developers and governments, and warned that unilateral shutdowns by one company would do little to improve safety.
Similar arrangements exist for other powerful technologies, but the company cautioned that building the necessary infrastructure and trust for them will take time.
“Not much longer left,” it said.