ChatGPT Has Been Handed the Right to Self-Rule. What Could Go Wrong?

Former British Prime Minister Rishi Sunak once thought artificial intelligence so risky that in 2023 he organized the world’s first “AI Safety Summit,” inviting policy makers and longtime AI doomer Elon Musk to talk up guardrails for the boom sparked by ChatGPT. Two years later and his view has softened considerably.

“The right thing to do here is not to regulate,” he told me last month at Bloomberg’s New Economy Forum, saying companies like OpenAI were “working really well” with security researchers in London who tested their models for potential harms. Those firms were volunteering to be audited. When I pointed out they might change their minds in the future, Sunak replied, “So far we haven’t reached that point, which is positive.” But what happens when we do?