Son says he spends two to three hours daily using ChatGPT and believes AI is already more knowledgeable than he is in many areas.

OpenAI disclosed that GPT-5.3-Codex contributed to its own development by identifying training issues and aiding deployment and evaluation, illustrating AI assisting in creation.

SoftBank’s Masayoshi Son says OpenAI’s latest model signals the possible emergence of artificial superintelligence, with AI potentially designing future models with minimal human input.

He argues that future AI designs may be produced by AI itself, accelerating beyond human capabilities, and in a CNBC interview suggested the next OpenAI model could be built by AI rather than human engineers because humans may no longer be smart enough to design it.

Anthropic contends that slowing AI development could improve safety, but warns that a slowdown may be impractical or could backfire if slower actors gain a competitive edge.

The discussion sits within broader debates on the pace of AI progress and safeguards as industries push for faster advances.

Anthropic cautions that fully autonomous self-improvement would shift human roles toward oversight, validation, and monitoring of AI-driven laboratories.

Anthropic warns that recursive self-improvement could accelerate AI progress, though it is neither achieved nor inevitable.

Experts stress the need for tools to validate, verify, and align AI behavior with human intent as capabilities advance.

Both SoftBank and Anthropic acknowledge benefits to science and healthcare from rapid AI progress, while cautioning about risks like reduced human control over AI systems.

Anthropic has urged governments and labs to prepare for a future where AI can design, build, and train models with minimal human involvement, calling for safeguards or a brake mechanism.

Anthropic describes a potential feedback loop where faster coding, debugging, and research by frontier models enable AI systems to autonomously build better models.