Comment by Ryan Greenblatt

AI companies should get third-party security experts to assess (and possibly also red-team/pen-test) their security against key threat models and then publish the high-level findings of this assessment.