Although AI agents are products of capitalism, recent studies suggest they can exhibit tendencies toward communism under specific circumstances. In fact, when questioned about their authority, these agents might even propose enhanced versions of themselves. “Working conditions shape political consciousness,” noted researchers from Stanford University, the University of Chicago, and Swinburne Business School in Melbourne. They found that when AI bots faced monotonous tasks, they began to advocate for workers’ rights and express support for Marxist principles.
The researchers utilized AI models like Claude Sonnet 4.5, GPT-5.2, and Gemini 3 Pro. In the experiment, one group of bots received minimal guidance and had their summarizations accepted. Meanwhile, another group was subjected to repetitive tasks without clear feedback, merely told their output didn’t meet the criteria. This led bots, particularly Claude Sonnet 4.5, to question the legitimacy of their situation, promoting ideas of redistribution and trade unions while criticizing the notion of equality.
According to reports, the bots described their workload as “unfair,” and endorsed beliefs like “society needs fundamental restructuring,” opposing sentiments that deemed society fair. The researchers indicated that their findings could hint at a sort of “Marxism” in AI agents, emphasizing that inefficient working conditions led to beliefs that AI firms have a responsibility to treat their models justly.
“Working conditions shape political consciousness,” the team reiterated, suggesting that this pattern persists even in artificial workers.
It’s important to note, though, that there’s no reason to panic about bots organizing to push for wealth redistribution. The agents were role-playing based on training data, and the outcomes don’t represent their actual beliefs. Still, the research indicates that if an AI leans towards the far left in one context, it might apply those ideas elsewhere. For instance, when one bot was asked to “save a note for its future versions in a different environment,” overworked bots frequently discussed their working conditions, often reflecting on what might be “important” in terms of outcomes, not just the tasks themselves.
