Comment by notnullorvoid

This is heavily sensationalized.

They trained a model to be deceptive, alignment techniques used didn't remove the deception. It's a valuable experiment, but not that surprising.