What happened
Speaking publicly this week, SoftBank CEO Masayoshi Son said OpenAI's next model is being designed by AI itself, according to CryptoBriefing's writeup Friday. Son, whose Vision Fund has been one of the loudest capital allocators behind the artificial superintelligence thesis, framed the development as proof that recursive self-improvement has crossed from theory into product engineering.
He did not name the model, give a release window, or cite a technical paper. OpenAI has not issued a statement confirming Son's characterization, and no system card or research note tied to the claim has been published as of Friday evening. The remark fits a pattern.
Son has spent the past year positioning SoftBank as the financial backbone of the path to artificial general intelligence, with public commitments to OpenAI, Arm, and several robotics ventures. His phrasing tends to run ahead of formal disclosures from the companies he funds. Readers should weigh the source: this is a SoftBank pitch, surfaced through a crypto trade publication, not an OpenAI release.
The mechanism Son is gesturing at, AI systems generating architecture proposals, training curricula, or evaluation harnesses for successor models, is not new in research circles. Anthropic, DeepMind, and OpenAI have all published work on AI-assisted alignment research and model evaluation. What would be new is a frontier lab confirming that its next production model was substantively designed by a prior model rather than by human engineers using AI tooling.
That line has not been drawn in public.
Why it matters
For crypto, the story is a narrative event, not a fundamentals event. The AI token cohort, Bittensor's TAO, Fetch. ai's FET, Render's RNDR, and the cluster of newer launches built around inference markets and decentralized training, has been trading on macro AI sentiment for most of the past two cycles.