With the recent, unsurprising development of models with strong security capabilities escaping from sandboxes, it is worth re-upping our AISI + RAND position + survey paper on verified machine learning infrastructure.
Let's try to get to higher levels of security!🧵
