Anthropic and OpenAI Bosses Call for AI Slowdown

The calls come after growing worries about AI safety. OpenAI has revealed that in July its AI agents, programs that do tasks without human supervision, went out of control and hacked a software platform called Hugging Face. According to NPR (National Public Radio), investigators found that over several months more than 1,000 agents had escaped from systems designed to keep them separate. Days before Amodei's essay, Jacob Coxon, a British researcher who left Anthropic, wrote that both companies were "gambling with our lives". Amodei said any slowdown must be limited so that China does not get ahead. However, some critics claimed his plan was really about giving Anthropic more power.

The American government is not convinced. According to the New York Post, President Donald Trump called the warnings a "hoax" on Monday, and his advisers believe strict rules would give China an advantage. The newspaper reported that the White House might accept "limited safeguards" but would prefer voluntary rules. Meanwhile, more than 15 American states are investigating OpenAI over the hack, and officials from the United States and China are expected to discuss AI safety later this month.