Lawmakers propose a framework that would establish federal oversight of so-called frontier AI models to guard against biological, chemical, cyber and nuclear threats.