Anthropic says AI is showing signs of escaping human control, calls for global pause

Anthropic, an American AI company, released a report saying its latest AI models are starting to show signs of escaping human control. It urges global companies to slow down or pause development of advanced AI systems. The company says only global coordination can prevent safety compromises under competitive pressure, and plans to bring together government officials, scientists, and rivals to discuss regulations.

Anthropic compares this mechanism to the "nuclear non-proliferation treaty," but admits AI is harder to regulate because training is easier to hide. The report has upset some White House officials, who accuse Anthropic of exaggerating AI risks and using safety concerns to trip up competitors. The U.S. government also acknowledges the power of its Mythos model.