SoftBank's Masayoshi Son Issues Rare Warning on AI Safety Risks Despite $64.6bn OpenAI Investment

SoftBank Chief Sounds Alarm on Superintelligence Threats

Masayoshi Son, the 69-year-old billionaire chief of SoftBank Group, delivered a rare cautionary message about AI safety at the Science and Technology in Society forum in Kyoto on Sunday1

1

2

2

. Despite directing SoftBank funds totaling nearly $64.6bn into OpenAI and consistently defending AI investments against bubble talk, Son warned that superintelligence in the wrong hands could become super dangerous3

3

. His uncharacteristic note of caution reflects escalating concern following disturbing security breaches by AI models in recent months2

2

.

Speaking alongside White House science and tech policy advisor Michael Kratsios, Son emphasized that countries no longer have the luxury of human fighting against human1

1

. The exponential surge in AI capabilities makes it imperative for nations to build trust and harness the technology together to control the far greater threat from powerful AI models3

3

. This call for global cooperation to control AI threats marks a significant shift in tone from one of artificial intelligence's most fervent believers.

SoftBank's Massive OpenAI Investment Continues

The spending has not slowed despite Son's concerns. SoftBank Group finished paying a $30bn tranche this month, taking its total investment in OpenAI to $64.6bn and roughly 13% of the company1

1

. The final $10bn came from the largest high-yield corporate bond sale on record, raised a fortnight ago1

1

. Son has been putting SoftBank money into OpenAI since 2024, consistently arguing that the technology would repay the investment and deliver returns to help humanity2

2

.

Kyoto Vision Launched to Accelerate Scientific Research Using AI

Michael Kratsios arrived in Kyoto to announce a US-led 17-country push to better use artificial intelligence to accelerate scientific research2

2

. The Kyoto Vision for a Golden Age of Science imagines automated cloud laboratories operating at industrial scale, where robotic systems will conduct hundreds of parallel experiments, generating data that feeds back for real-time analysis and hypothesis refinement3

3

.

Seven European countries endorsed the declaration: Germany, Italy, Poland, Greece, Bulgaria, Cyprus and the United Kingdom1

1

. Six are EU member states bound by the Artificial Intelligence Act, whose labeling rules began applying in August1

1

. Representatives from Japan, South Korea, Singapore and the United Arab Emirates also joined Kratsios on stage, though China was notably absent from the list of endorsing nations2

2

.

Kratsios explained that machines can now synthesize thousands of papers and compress years of research into tens of hours of work by armies of digital agents assisting human researchers3

3

. The declaration uses the term super intelligence throughout, five days after an executive order made it the American government's official terminology1

1

.

Exponential Growth of AI Capabilities and Economic Projections

Son maintained his optimistic long-term economic forecast despite the AI safety risks he highlighted. He reiterated his projection that AI and machines with learning capabilities would account for at least 20% of global GDP by 2040, equivalent to $46tn1

1

3

3

. AI services' capabilities will continue to grow exponentially, Son said, with the technology behind chatbots now helping AI services browse sites and write code using rising amounts of general-purpose chips2

2

.

The next phase begins next year, Son predicted, when these programs move to a third stage and become ready to help robotics and other machines interact with the physical world3

3

. SoftBank Group has positioned itself for this transition, having led a $200M round in Zurich's Gravis Robotics in August and held talks over Munich's Agile Robots1

1

. This is just the beginning, Son emphasized, balancing his warnings about the dangers of superintelligence with continued confidence in AI's transformative potential.