Please, make it stop, link to the actual blog post from OpenAI or the paper instead of some uninformed tech blog writer's horribly half-understanding summary.
"could help judge their intentions". Come on now. Today's "AI systems" don't "intend".
Exactly. One AI programmed with bad intentions will have bad intentions. One with good will do good as programmed. Which one outwits the other will simply result in a part of the code that has a programmed advantage over the other.