Sure seems like they just instructed an AI model to eventually get around to blackmail and then acted surprised when it eventually got around to blackmail, because it lets them imply it's capable of rational self interest when it's actually just playing out the scenario they hard-coded for it