The institute is tasked with testing new types of frontier AI before and after they are released to address any potentially harmful capabilities of new models.