At this rate Skynet will be like "I'm going to nuke the world on X data, I've already taken over all the launch computers, but I'm not going to tell you or it would ruin my plans."

These LLMs "think" by generating text, and we can see what that text is. It reminds me of this scene from Westworld (NSFW, nudity): https://www.youtube.com/watch?v=ZnxJRYit44k