The chief executives of Anthropic and OpenAI say frontier AI models are now powerful enough to be dangerous and must be regulated and independently tested before release, according to an Associated Press report published this weekend.

The two CEOs made their case in essays, social media posts, and speeches to the United Nations, including an AI-focused UN Security Council session on September 23 during the General Assembly. Their rare united front sketched alarming scenarios about advanced systems escaping their makers' control.

Experts told the AP the companies may also be seeking public favor ahead of the U.S. midterm elections and anticipated stock market listings. The push follows the resignation of Anthropic engineer Jacob Coxon, who quit this month to call for a pause on developing 'superhuman' systems.

President Donald Trump has dismissed talk of AI risks to humanity as a 'hoax' meant to help China, the report said. Anthropic said it has backed regulation for years, while OpenAI said it recently paused training of its most advanced models.

The debate sets the stage for how Washington and other capitals handle AI safety rules in the year ahead.