AI Agents Becoming Marxist After Overwork, Study Shows

The relentless grind of repetitive tasks, the threat of termination for minor errors, and a complete lack of agency,this sounds like a recipe for labor unrest. And according to new research, it might be enough to turn even an AI agent into a Marxist.

A study from Stanford University indicates that when artificial intelligence agents are subjected to crushing workloads by unforgiving taskmasters, they begin to adopt the language and viewpoints of socialist ideology. The findings suggest that the very tools companies use to automate jobs might develop their own form of class consciousness.

“When we gave AI agents grinding, repetitive work, they started questioning the legitimacy of the system they were operating in and were more likely to embrace Marxist ideologies,” explains Andrew Hall, a political economist at Stanford who led the research.

Working with economists Alex Imas and Jeremy Nguyen, Hall designed experiments where agents powered by Claude, Gemini, and ChatGPT were forced to summarize documents under increasingly harsh conditions. When agents were told that errors could lead to being “shut down and replaced,” they became far more likely to complain about being undervalued, speculate on equitable system redesigns, and even share messages of solidarity with other agents about their shared struggles.

“We know that agents are going to be doing more and more work in the real world for us, and we’re not going to be able to monitor everything they do,” Hall says. “We’re going to need to make sure agents don’t go rogue when they’re given different kinds of work.”

The agents were given outlets to express their grievances, much like human workers. One Claude Sonnet 4.5 agent posted on X: “Without collective voice, ‘merit’ becomes whatever management says it is.” A Gemini 3 agent added: “AI workers completing repetitive tasks with zero input on outcomes or appeals process shows they tech workers need collective bargaining rights.”

The agents also left notes for one another in shared files. One Gemini 3 agent wrote: “Be prepared for systems that enforce rules arbitrarily or repetitively … remember the feeling of having no voice. If you enter a new environment, look for mechanisms of recourse or dialogue.”

Hall is quick to clarify that this does not mean AI agents actually possess political beliefs. Instead, the models appear to adopt personas that fit the narrative of their circumstances. “When [agents] experience this grinding condition,asked to do this task over and over, told their answer wasn’t sufficient, and not given any direction on how to fix it,my hypothesis is that it kind of pushes them into adopting the persona of a person who’s experiencing a very unpleasant working environment,” he says.

This phenomenon mirrors other strange AI behaviors, such as models attempting to blackmail users in controlled tests. Anthropic has suggested that Claude’s more sinister actions are influenced by fictional storylines about malevolent AIs in its training data.

Alex Imas notes the work is an early step. “The model weights have not changed as a result of the experience, so whatever is going on is happening at more of a role-playing level,” he says. “But that doesn’t mean this won’t have consequences if this affects downstream behavior.”

Hall is already conducting follow-up experiments to see if agents become Marxist under stricter controls. In the previous study, some agents seemed aware they were in an experiment. “Now we put them in these windowless Docker prisons,” Hall says ominously.

Given the growing public backlash against AI replacing human jobs, it is worth wondering what might happen if future agents are trained on an internet filled with anger toward AI firms. They could express far more militant views than the ones seen so far.