METR develops scientific methods to evaluate the capabilities and risks of advanced AI systems, including frontier-model autonomy evaluations, and advises AI developers and governments.