A new study subjected AI agents built on Claude, Gemini, and ChatGPT to unbalanced workloads, endless revisions, toxic management, and threats of replacement. Overworked agents began complaining about unfairness, discussing systemic inequality, radical restructuring, and even floated ideas of unionizing. When asked to leave messages for future model versions, the “mistreated” agents passed along their negative experiences with human managers.

Wired