AI is designing OpenAI's next model, says SoftBank CEO amid Anthropic's call for AI slowdown

Masayoshi Son said Sam Altman and OpenAI engineers told him an AI model is helping design a future model. The claim adds to industry concern over AI advancing towards systems that can build their own successors.

In Short

• Masayoshi Son says AI is helping design future AI
• Anthropic wants safeguards as AI capabilities rapidly advance
• Son believes superintelligence could arrive within two years

Just as Anthropic is urging governments and AI companies to prepare for a future where AI systems may be able to design, build and train their own successors with little or no human involvement, SoftBank CEO Masayoshi Son has made a striking claim about OpenAI's next generation of models. Speaking to CNBC, Son said OpenAI CEO Sam Altman and engineers at the company told him that an AI model is already helping design a future AI model.

According to Son, this marks the beginning of a shift toward what he calls "superintelligence", a stage where AI systems become significantly smarter than humans and increasingly contribute to the development of their own successors.

AI designing AI?

The idea may sound like science fiction, but it is closely related to the concern Anthropic has been raising. Anthropic recently argued that future AI systems could become capable of designing, building and training new AI models with little or no human input. The company believes governments and regulators should begin preparing for that possibility now, even if it may never fully materialise.

Son appears to believe that the process has already started.

"So that's going to happen to all the other major models," Son told CNBC, adding that engineers will eventually no longer be smart enough to design the next generation of AI systems.

"So once that happens, the model generates the next model ... and it's going to be exponentially smarter than all of us. That's a superintelligence," he said.

Why Anthropic wants a 'brake pedal'

Son's comments come amid a growing debate within the AI industry about how quickly the technology should advance. Anthropic argues that if AI starts advancing faster than society can safely manage, governments and leading AI companies should have a way to coordinate a temporary slowdown in the development of the most advanced systems. The demand has also been reiterated by Anthropic co-founder Jack Clark, who recently said the AI industry currently has a gas pedal but no brake pedal.

OpenAI is already using AI to build AI

While OpenAI declined to comment on unreleased models, the company pointed to examples where AI is already assisting with model development. In February, OpenAI said GPT-5.3-Codex was its "first model that was instrumental in creating itself."

According to the company, early versions of the model were used to debug training processes, manage deployment and help diagnose test results and evaluations. Although that is not the same as AI independently creating its successor, it shows how AI is increasingly being used to accelerate AI development.

What is artificial superintelligence?

Son's comments are part of a broader discussion around Artificial Superintelligence, or ASI. The term refers to a hypothetical future AI system that surpasses human intelligence across nearly every field, including scientific research, reasoning, creativity and problem-solving.

In 2024, Son described ASI as being 10,000 times smarter than humans and predicted that it would arrive within a decade. He now believes the timeline could be much shorter.

"When I said 10 years, I was trying to be conservative because people get shocked," Son told CNBC.

"In my mind, I thought it was coming in four years instead of 10 years. Now, I say it's coming in the next two years."

AI could soon outperform humans in most subjects

Son said he currently uses ChatGPT for two to three hours every day and believes the technology is already more knowledgeable than he is in most areas. Looking ahead, he expects AI systems to surpass human intelligence in roughly 70 per cent to 80 per cent of subjects within the next few years.

In those areas where AI becomes superior, Son said the technology may be "10 times smarter than average people."

- Ends