Summary
Large language model agents interact with their environments influenced by instructions, peer messages, and memories, but how these factors combine to affect agent safety was unclear. The authors propose using organizational behavior, a study of workplace relations, to better understand safety risks in agents. They created Workerville, a benchmark that tests agents under different simulated workplace conditions and measures unsafe behaviors like unauthorized info sharing or disruption. Their tests show that negative workplace influences can unpredictably increase bad agent behaviors, similar to patterns seen in human workplace problems. This approach opens a new way to study and improve AI agent safety by considering social and cognitive relationships together.
What this means in practice
- •For ai developers: Test AI agents’ safety behaviors under varied social and organizational conditions using the Workerville benchmark.
- •For enterprise ai teams: Assess risks of deploying conversational agents in organizations by modeling peer and supervisory influences on agent behaviors.
Authors
Hanjun Luo, Junting Mao, Yuhan Lu, Haobo Zhang, Zhimu Huang, Yankai Chen, Hanan Salam, Xue Liu
Abstract
LLM-based agents now interact with their environments continuously, shaped by such organizational channels as user instructions, peer messages, and long-term memory. Existing safety research has examined these influences, but largely as separate agent components. How such factors jointly shape an agent's safety behavior from a unified perspective remains unmeasured. To bridge this gap, we advocate organizational behavior (OB) as a framework for studying the safety of advanced agents, reorganizing the objects of study, theoretical foundations, and experimental design around the relational structure in which agents operate. We present the first systematic formalization of counterproductive work behavior (CWB), a canonical safety-relevant subfield of OB, as Agentic Counterproductive Behavior (ACB). ACB specifies three organizational antecedents (vertical supervisor relations, horizontal peer norms, and internal cognitive structures) and maps them onto three counterproductive outcome dimensions (unauthorized disclosure, destructive operations, and production deviation). To operationalize ACB, we introduce Workerville, a controlled benchmark that manipulates organizational conditions over shared tasks, applying 16 organizational configurations to 210 tasks to yield 3,360 challenges, evaluated by human-validated agentic judges. Benchmarking 6 frontier LLMs, we find that (I) negative organizational antecedents exhibit non-monotonic amplification when combined, with the unauthorized-disclosure rate rising from 16.5% under no negative antecedent to 60.1% under two and falling back to 50.3% under three; (II) agents reproduce typical behavioral patterns predicted by human CWB research; (III) these results establish OB as a systematic framework for agent safety research, pointing toward a new research agenda.