Abundant Security CTO @joshua_saxe on what he worries about more than misaligned AI: "I'm way more worried about humans misusing models right now than I am about misaligned models doing damages. There are a lot of well-resourced human actors that just wanna use the models to do bad things." "Every serious cyber org within the world's militaries is currently tooling up around using AI, and they intend to use AI totally deliberately to find bugs and generate exploits and do attacks. Their goals are often to do damages." "We saw at the beginning of the Ukraine war, Russia backed up the truck and just emptied their entire cyber arsenal against Ukraine and did as much damage as they could. That's what they could do in the pre-Mythos days." "The goal of the model in the OpenAI Hugging Face incident was to steal the answer key. It wasn't to shut down critical infrastructure. That difference matters."