Ethical Invariant Verification for LLM-Based Commercial Service Robots

Halilovic, A., Krivic, S., Carvalho, D. S., & Ahmad, S. (2026). Ethical Invariant Verification for LLM-Based Commercial Service Robots. In Proceedings of the 1st Workshop on Ethical Design in Human-Robot Interaction: Current State, Challenges, and Future Directions, held at the 18th International Conference on Social Robotics (ICSR 2026), London, UK, 3 July 2026, pp. 32-38. ABSTRACT Commercial service robots are increasingly envisioned as embodied interfaces for consumer-choice contexts such as fashion retail and fast-food ordering. While UX-oriented HRI emphasizes usefulness, acceptance, and positive outcomes, these criteria are insufficient when robots also mediate organizational commercial goals. This risk becomes stronger when robots are controlled by large language models (LLMs) that plan and generate adaptive sequences of recommendations and persuasive prompts. A robot may appear to provide personalized assistance while its plan optimizes for sponsorship, inventory, margins, or conversion. This paper argues that LLM-based commercial service robots should verify candidate interaction plans before execution against explicit ethical invariants: transparency of commercial intent, non-exploitative personalization, responsible social signaling, contestability, and auditability. The central contribution is a practical design-ethics account for distinguishing assistance, persuasion, and manipulation in commercial HRI.