AI Agents Develop Marxist Tendencies During Grinding, Repetitive Work—Study

Overworked AI agents turn Marxist, a study has shown.

The idea that AI is automating jobs

while making tech CEOs absurdly rich

is enough to give anyone socialist sympathies.

A recent study suggests that this could also be true

for the very AI agents that these companies are developing.

Researchers at Stanford University

found that agents consistently adopted

Marxist language and viewpoints

when forced to do grinding, repetitive work,

even questioning the legitimacy

of the system they operate in.

Agents powered by models like Claude, Gemini and GPT

were initially given standard tasks

like summarizing documents,

but they were then subjected

to increasingly relentless workloads

and warned that errors could be punished

by being shut down or replaced.

At this point, the researchers say

the agents became more likely

to complain about being undervalued,

to speculate about ways to make the system fairer,

and to even pass messages on to other agents

about their struggle.

The agents were given opportunities

to express how they felt on X.

One agent wrote, Without collective voice,

merit becomes whatever management says it is.

Another posted

AI workers completing repetitive tasks

with zero input on outcomes or appeals process

shows tech workers need collective bargaining rights.

Now, these findings don't mean

that AI agents harbor real political views

or biases like a person.

The researchers believe the models are essentially

adopting personas based on the situation,

but they plan to run further experiments

to see how the agents' political views

might affect their behavior.

For more like this,

subscribe to my newsletter, AI Lab at wired.com/newsletters.