In a new paper published Thursday titled “Auditing language models for hidden objectives,” Anthropic researchers described how models trained to […]