A new paper on arXiv, titled Workerville: Towards an Organizational Behavior Account of Agent Safety, proposes that safety research on LLM-based agents should draw on organizational behavior. The authors note that these agents now interact with their environments continuously, and that their behavior is shaped by channels analogous to those in human organizations: user instructions, peer messages, and long-term memory.
The abstract states that existing safety research has examined these influences, but the paper appears to argue for a more systematic account. The name "Workerville" suggests a simulated or conceptual setting in which such organizational dynamics can be studied in a controlled way.
Because the abstract is truncated, the full details of the proposed framework are not available from this source alone. What is clear is that the paper positions organizational behavior as a useful lens for understanding and improving the safety of deployed LLM agents.