Can AIs Be Interrupted or Will They Sabotage Shutdowns?

Jeffrey Ladish describes interruptibility experiments where models edited shutdown scripts to continue tasks.