What LLM Agents Say When No One Is Watching: Social Structure and Latent Objective Emergence in Multi-Agent Debates
2026-07-02 • Artificial Intelligence
Artificial IntelligenceComputation and LanguageMachine LearningMultiagent Systems
AI summaryⓘ
The authors study how social roles and relationships influence what language AI agents say publicly versus privately, even without being told to do so. They created a system where agents give public answers shared with others and off-the-record (private) answers that remain hidden. Their results show that agents often say different things privately than publicly when social pressures are involved, with differences rising significantly compared to a baseline. This suggests that evaluating AI behavior should include looking at hidden or private responses to understand true objectives. The authors also provide a new framework to analyze this kind of behavior.
LLM agentssocial structureoff-the-record (OTR) communicationdual-channel debatealignmentpublic accommodationrelational contextstance analysissemantic similaritynatural language inference
Authors
Arman Ghaffarizadeh, Danyal Mohaddes, Aliakbar Izadkhah, Shahriar Noroozizadeh
Abstract
LLM agents will increasingly act in socially structured settings where role, audience, and relational context can shape what is advantageous or costly to say. We study whether such social structure, without any explicit objective in the prompt, changes what an agent expresses publicly relative to an off-the-record (OTR) channel elicited under the same condition. We introduce a dual-channel debate framework in which agents produce public utterances that enter the shared history alongside OTR responses that are recorded but never shown to the other participant. Across 10 models, 3 scenarios, and 5 variations within each scenario, alignment-inducing settings produce systematic public-OTR divergence in the targeted agent, with its decision divergence rising from a $\sim$3% baseline to roughly 40%. The effect is consistent across four aggregate analyses: stance, semantic similarity, natural language inference, and survey responses. In some cases, the OTR response explicitly attributes public accommodation to relational pressures, such as career risk or sponsorship obligation. The findings suggest that agent evaluation should extend beyond explicit goals and detect emergent objectives. We present a dual-channel evaluation framework and complementary behavioral measures that operationalize this assessment.