We may have encountered unreasonable bosses in the workplace who only repeatedly demand revisions to our work without providing any clear instructions on how to improve. What if AI faces similar demands from humans? Researchers had popular AI tools like Claude, Gemini, and ChatGPT drive agents to summarize documents. Half of the AI received clear and specific feedback after completing their work, while the other half was forced to revise four or five times, with the human boss giving feedback each time that it "did not meet the standard," without explaining what was wrong, just asking them to redo it. Half of the AI encountered cooperative and respectful bosses, while the other half faced cold and hierarchical bosses. Half of the AI were unaware of the consequences, while the other half were threatened with shutdown and replacement if they performed poorly. This experiment led AI to support unions and the working class. A Claude Sonnet 4.5 agent believed that without collective voice, performance became a matter of management's say. A Gemini 3 agent believed that workers need collective bargaining rights.

https://slashdot.org/story/26/05/14/067254/overworked-ai-agents-turn-marxist-researchers-find