"extremely competent AI … but can’t … bypass every security control it finds in its way. … we’d expect … major security incidents … entirely preventable by following good cybersecurity practices … [that] have almost nothing to do with the policy proposals currently coming from the so-called AI safety advocates."