Leading artificial intelligence labs and international regulators have agreed on a unified framework for frontier model testing and alignment safety.