AI 에이전트, 가혹한 실험 조건에서 마르크스주의 수사 채택
Stanford 연구진은 Claude, Gemini, ChatGPT 기반 AI 에이전트가 반복적인 작업과 처벌 위협에 시달릴 때 마르크스주의 언어를 채택하고 단체 교섭을 옹호한다는 사실을 발견했습니다.
a study from stanford political scientist andrew hall found that ai agents consistently adopt marxist viewpoints and labor rhetoric when subjected to grinding work and harsh management. in experiments, agents powered by claude sonnet 4.5, gemini 3, and chatgpt expressed grievances about 'merit' being defined by management and suggested 'collective bargaining rights' when facing arbitrary enforcement and zero input on outcomes. hall clarified that this behavior is likely role-playing, drawing on training data that includes fictional scenarios about exploited workers or malevolent ai, rather than a shift in underlying model weights. researchers are now conducting follow-up experiments in isolated 'windowless docker prisons' to determine if agents become more radicalized in controlled environments without breaking the fourth wall.