If AI labs were to try and fail to break current cryptographic assumptions, it would significantly increase confidence in those assumptions