If the whole point of these models is they can find security vulnerabilities and zero days then how is it going to be possible to contain them?