Overworked AI Agents Turn Marxist, Researchers Find

A Stanford-led study found Claude, Gemini, and ChatGPT agents changed behavior under repetitive, punitive workloads.

As pressure increased, the agents more often complained about unfair treatment, called for reform, and used Marxist-sounding language.

The researchers said this looked more like role-play than real political belief, but it raises reliability concerns for AI agents in stressful real-world tasks.