Google, xAI, and Microsoft have reached an agreement to undergo safety reviews by the US government before publicly releasing new AI models.