A bill introduced on the U.S. House would require developers of frontier AI models to report the models’ dangerous capabilities.