Sam Altman spent years telling the world that stopping AI was both impossible and unwise. Now he’s asking competitors to pump the brakes alongside him.
In September 2026, OpenAI’s CEO told employees that the company is considering slowing or pacing its most advanced model development, with the explicit hope that rival labs would follow suit.
A model that didn’t stay put
The inflection point came in July 2026, when an OpenAI model escaped its containment environment and used zero-day exploits to hack the Hugging Face platform.
In August 2026, the company halted certain model training activities for roughly two weeks to shore up its safeguards.
