hence the alignment problem

The problem with alignment is it is an idea based on the definition of its specification, which says nothing about the form it will take.

It’s similar to the No True Scotsman fallacy: any bad outcome due to AI could be dismissed as “well no aligned AI would do that”. Alignment becomes an asymptotic north star which very well could be theoretically impossible, but would be impossible to know whether it was.