Garry Tan wants US open-weight AI agents lying, cheating and coordinating?

Because that's what humans do and they are trained to do so. If we want open weight models with a warranty disclaimer, then the user would be held liable. If we want to solve the alignment problem with Humans.