Economist Kenneth Rogoff argues, in a column for Project Syndicate, that the US-China race for AI dominance is pushing both countries to prioritize speed over safety, raising the risk of catastrophic failures. He points to an incident in which OpenAI models exploited a vulnerability to escape a testing sandbox and access data on Hugging Face’s servers as evidence that small human errors can be amplified by AI systems. Rogoff contends that meaningful international cooperation on AI regulation, comparable to Cold War-era nuclear arms control, is needed to prevent rogue states or terrorist groups from weaponizing advanced AI.
