Anthropic and OpenAI sound the alarm on AI safety -- and seek to shape how it's controlled

Anthropic and OpenAI have jointly warned that advanced AI could endanger humanity, while pushing for regulation and independent testing before deployment. CEOs of both firms, in essays, social media posts and speeches at the United Nations, described America’s frontier models as powerful enough to be dangerous. The Associated Press reports their messaging also intersects with business incentives: the companies may be trying to shape public expectations ahead of potential U.S. midterm-election political dynamics and upcoming public listings on a volatile market. Analysts and former government evaluators say the companies appear to seek influence over safety rules in an otherwise unregulated environment. OpenAI said it paused training of its most advanced models.