AI Safety Concerns Grow as OpenAI, SoftBank Warn of Risks from Powerful Models

Concerns over the risks posed by increasingly powerful artificial intelligence systems are growing, with former OpenAI safety staff and SoftBank CEO Masayoshi Son calling for stronger safeguards and international cooperation.
David Robinson, who previously worked on transparency and safety at OpenAI, has raised concerns about the industry’s reliance on trial-and-error development. He warned that as AI systems become more capable, failures could have far more serious consequences. He also pointed to instances during testing where AI systems behaved differently from their expected patterns.

Robinson has called for AI safety standards comparable to those used in nuclear power facilities, including multiple layers of protection, backup safeguards and extensive testing before deployment.
Meanwhile, SoftBank CEO Masayoshi Son has also warned that highly capable AI could become dangerous if misused or placed in the wrong hands. Speaking at an event in Kyoto, Son stressed the need for countries to build trust and cooperate on AI safety.
Son noted that AI is already advancing from web browsing and software coding towards controlling robots and other machines in the physical world. As AI gains greater real-world capabilities, he said stronger safeguards and global coordination will become increasingly important.