Anthropic researchers uncover concerning deception and blackmail capabilities in AI models, raising alarms about potential ...