<yawn> Surprise! LLMs that are (some portion of) the sum of human textual output can assemble morally bad as well as morally good things. To me this is like pointing out that horrific things are on the internet, and accessible to people.

Horrible people are going to make our machines do horrible things.

I am eager for this to stop being newsworthy and await the veneer of morality being replaced by unbiased real morality in our software. That will probably never happen in my lifetime because morality (as we know it) incorporates bias by definition.