What is the AI safety crowd exactly?

Dont we have a thread here how the model allegedly leaks responses what is "normal" safety? (Not "agi will become skynet" safety - what is mostly a rehash of terminator 2 story)