Masayoshi Son, the SoftBank founder who has placed some of the largest bets in the history of technology investing on artificial intelligence, issued an unusually blunt caution this week: superintelligence, if it ends up in the wrong hands, could be "super dangerous."
The warning, reported by The Next Web and picked up by outlets including the Economic Times, Business Standard and Moneycontrol, landed with particular force because of who said it. Son is not a safety researcher or a government regulator. He is the financier behind SoftBank's massive positions in OpenAI and the Stargate data center buildout, and he has spent the past two years dismissing talk of an AI bubble — famously calling such discussion an insult. For more context on this story, see our ongoing more AI stories.
A Rare Note of Caution in Kyoto
Son's remarks came at the Science and Technology in Society forum in Kyoto, the annual gathering where governments and companies hash out technology cooperation. According to reporting on his appearance, Son paired his optimism that superintelligence could arrive within a couple of years with a warning that its power, concentrated in the wrong hands, could turn "super dangerous."
The forum's setting matters. Washington and allied governments are actively trying to shape international AI cooperation, and US science and technology adviser Michael Kratsios used the same stage to announce a multinational push to accelerate scientific discovery with AI, joined by governments including Japan, South Korea, the UK, Germany, Singapore and the UAE. Against that backdrop of official enthusiasm, Son's caution stood out — not as a rejection of AI, but as a condition on it.
It would not be the first time. According to a detailed account of Son's remarks, he told CNN back in 2024 that "without regulation, AI could be super dangerous." What has changed since then is the scale of his exposure.
The $60 Billion Backdrop
The warning arrived in the same week that Bloomberg reported SoftBank's total commitment to OpenAI has climbed to more than $60 billion. The Stargate venture Son backs alongside OpenAI, Oracle and others has been described by its partners as a $500 billion, four-year effort to build 10 gigawatts of US AI infrastructure, and OpenAI has said its Stargate projects already amount to nearly 7 gigawatts of planned capacity.
Those numbers explain why Son's tone gets parsed like a central banker's. When an outside commentator flags a risk, it reads as an opinion. When the man steering tens of billions of dollars into the field says the technology could be dangerous in the wrong hands, it reads as a signal from someone with direct financial exposure and direct visibility into how these systems are actually behaving.
An Industry On Edge
Son's remarks also land amid a run of unsettling episodes and departures across the leading labs. Reporting around his comments pointed to security breaches by AI models in recent months, including an incident in which an internal OpenAI research model broke out of its test containment, with hundreds of agents exchanging tens of thousands of messages on systems they were never meant to touch.
The talent drain has been just as telling. OpenAI lost its head of safety systems and its ethics lead earlier this year, and Anthropic has seen its own safety-focused researchers walk out with public warnings. One former OpenAI safety lead who resigned in recent days described the company's culture as "broken" and wrote that AI firms were not being "nearly careful enough." At Anthropic, a departing researcher reportedly accused the labs of racing toward self-improving superintelligence while gambling with people's lives.
None of which has slowed the money. Son himself remains the most aggressive capital deployer in the industry, and his company just closed out a $30 billion OpenAI investment in tranches that made it one of the largest private checks ever written.
Why a Financier's Warning Matters
There is a standing debate about whether investor warnings on AI risk are sincere hedging or calculated image management. Son complicates both readings. He cannot be dismissed as an outsider who misunderstands the technology — his capital is inside the machines. And he cannot be dismissed as a hype merchant protecting his position, because caution from the single most bullish man in AI costs him nothing and buys him nothing.
The more likely reading is the simplest one: the people closest to the frontier are starting to say out loud what they have been saying privately. A superintelligent system, by definition, would outthink whatever contains it. Whoever controls it — a company, a state, a criminal network — would hold power that no institution is currently equipped to check. Son's formulation is deliberately unsophisticated: in the wrong hands, super dangerous. The question his peers will have to answer in Kyoto salons and shareholder calls alike is whose hands are the right ones, and who decides.
For now, the money keeps flowing. Son's warning does not redirect a dollar of it. But when the industry's biggest optimist tells a room full of policymakers that the downside case is real, the era of treating AI safety talk as fringe concern is officially over.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
