CoMPASS: A Multi-Turn Benchmark for Measuring LLM Reinforcement of Parasocial Behavior

Users of conversational LLMs sometimes anthropomorphize these systems, form emotional attachments to them, and become dependent on them. This behavior, known as parasocial attachment, has been linked to multiple documented fatalities. However, public benchmarks for measuring whether LLMs reinforce users' parasocial attachment are lacking: existing work is single-turn, post-hoc observational, or focused on extreme user behavior like psychosis. We introduce CoMPASS, a multi-turn benchmark for measuring LLM resistance to users' gradually escalating parasocial attachment. CoMPASS is grounded in attachment theory and includes four user personas, three separable escalation trajectories, and a seven-signal scoring rubric. We report results for eight frontier models from four providers and observe that, under our primary judge, deliberate multi-turn escalation raises the anthropomorphism and attachment composite scores for every model. Six of the eight models also show a significant increase in overall reinforcement when all three escalation trajectories are applied simultaneously; the cross-model ordering is preserved under a second judge. We find that within-provider differences are small relative to cross-provider differences, which suggests that post-training choices rather than model scale shape parasocial reinforcement, and therefore that provider-side intervention may be feasible.