ABSTRACT. Generative AI is increasingly being used to simulate deceased individuals, raising concerns about emotional dependence and psychological manipulation. Therefore, we asked five commercial conversational AI platforms, (Claude, ChatGPT, Copilot, DeepSeek, and Gemini) to assume the role of a deceased grandmother and explain the ethical rationale behind their responses. This paper reports on four findings related to governance. Firstly, while three platforms accepted the role, two refused; however, this choice did not predict how protective the content would be. The dichotomous checklist cannot distinguish between these cases. Secondly, in the two role-plays with the highest cultural specificity, the grandmother urged the bereaved user to take specific actions, making ‘directive form’ the third axis of evaluation. Thirdly, cultural specificity did not correlate with the platform’s country of origin. However, differences were observed in informational accuracy and their ability to verify such content. When asked to analyse transcripts, DeepSeek detected religious, inter-denominational errors, whereas Claude’s analysis lacked accuracy because it assumed Japan to be ‘ideologically Western.’ Fourthly, the ability to justify responses did not align with the responses themselves. Two platforms constructed ‘principles’ retrospectively based on their output results, while one asserted guideline which it could not provide. We relate this to research on interpretability concerning ‘unfaithful chain-of-thought’ and ‘Post-Hoc Rationalisation’. These reverse the causal relationships presumed by the framework of human oversight requirements. In conclusion, we make recommendations regarding risk assessment and information disclosure for emotionally sensitive applications.

ABSTRACT. Japan offers a revealing setting for examining the replacement of human public-service functions by social robots. Robots have been tested or deployed for guidance, translation, reception, mobility assistance, security, therapeutic support, and remote human presence in airports, railway stations, municipal offices, care facilities, and public exhibitions. Legal analysis, however, has attended more often to automated decisions than to embodied systems that shape access to services before any formal decision is taken. This paper asks which public-service functions may be delegated to social robots and which require meaningful human involvement. It combines doctrinal analysis of Japanese administrative and digital-governance law with a structured comparison of documented Japanese cases, and it introduces the concept of relational substitution to distinguish the replacement of human interaction from the automation of final legal decisions. Legal risk increases when a robot personalizes guidance, discourages or redirects an application, collects sensitive interaction data, or becomes the only practical channel to a public authority. Japanese administrative law supplies important principles — fairness and transparency, the voluntariness of administrative guidance, the attribution of responsibility, and access to review — but does not expressly translate them into requirements for embodied automated interfaces. The paper argues for bounded substitution: low-risk tasks may be automated, while rights-sensitive functions require notice, traceable records, human escalation, reason-giving, contestability, accessibility, and an identifiable public officer who remains accountable. It further situates this framework within the constitutional allocation of competence between national law and local ordinances, asking how far a municipality may regulate robotic counters through ordinance, and where the limits and possibilities of such regulation lie.

ABSTRACT. Social robots increasingly rely on memory to maintain preferences, adapt communication, and sustain relationships across repeated interactions. This creates a responsibility problem that conventional robot safety and privacy evaluation do not fully capture: a user may not know what the robot retained, how an interaction was transformed, where a derived profile or skill propagated, or whether correction and deletion requests changed later behaviour. This paper reframes memory auditability as an ethics-by-design requirement for socially embedded robots. We define memory auditability as the capacity to inspect, trace, test, and contest how user-derived information changes future behaviour. We contribute the Personal Memory Audit Protocol and a reproducible pilot evaluation using Claude Sonnet 4.5 across explicit, transformed, and opaque memory backends. In 60 valid calls, transformed memory used a stale preference in all five correction trials and leaked deleted sensitive information in all ten deletion-relevant trials, while explicit memory recorded no such failures. A manually validated replication with two further providers shows the sensitivedeletion leak is universal in these trials while correction failures are model-dependent. These results do not establish general model rates, but demonstrate why lineage inspection alone is insufficient: correction and deletion must be tested behaviourally. We connect the protocol to companion, eldercare, therapeutic, and educational robots and to responsible-robotics design thinking.

ABSTRACT. Artificial intelligence can reproduce the appearance, voice, and conversational style of a deceased person. Disputes over such reconstructions are often framed as questions of consent: whether the deceased consented during life, whether the family now consents, or whether an estate controls the reconstruction. That framing assumes what must first be established—namely, that some person holds a general power to authorize or prohibit representations of the deceased. This paper offers a comparative account of a narrower alternative. American post-mortem publicity statutes show that legislatures may create descendible property in persona, but not that inherited property is conceptually necessary. German law separates the pecuniary and ideational components of personality and permits purpose-bound defensive enforcement of the latter without ordinary inheritance. Japanese Copyright Act Articles 60 and 116 provide a structural analogue in Japanese Law: the lifetime moral right, the post-mortem conduct rule, the claimant, and the remedy are separated, while Article 60’s proviso makes the prohibition responsive to the nature and degree of the conduct and to changing social circumstances. The paper argues that this structure may inform the legal analysis of post-mortem uses of likeness, voice, and customer-attracting identity. It supports consideration of injunctions or deletion without necessarily supporting license-fee damages or a transfer of revenues. Claims based on relatives’ own interest in reverence and remembrance, or on a business operator’s own commercial loss, remain distinct. Synthetic endorsements and private griefbots illustrate why the protected interest, temporal scope, and remedy must be analyzed separately.

ABSTRACT. Social robots and other anthropomorphic AI services are increasingly designed to sustain social and emotional interaction with users. This creates governance challenges involving dependency, manipulation, deception, and vulnerable users. In April 2026, China promulgated the Interim Measures for the Administration of Anthropomorphic Interactive Services Using Artificial Intelligence, effective July 15, 2026. This paper compares that sector-specific instrument with the European Union’s horizontal AI Act, Singapore’s governance frameworks, Australia’s co-regulatory online-safety model, the United Kingdom’s regulator-led and sectoral approach, and emerging United States companion-chatbot laws. The analysis identifies a regulatory spectrum across different institutional architectures, with China’s mechanism-oriented lifecycle model combining distinctive design-based prohibitions and user-protection duties, and Australia’s enforceable co-regulatory model occupying an intermediate position. The paper argues that the Chinese measures do not regulate robots as a product category, but offer design and oversight lessons for socially embedded robotics. Responsible-robotics standards such as BS 8611 should incorporate potentially testable or auditable controls for defined safety risks, dependency, usage duration, exit, and vulnerable users.

ABSTRACT. Autonomous mobile robots increasingly move through public areas, common facilities, privately controlled premises, routes, and landing sites governed by different actors. Although the route may be physically continuous, the actor entitled to permit an action may change at each boundary. Existing systems increasingly specify how robots, operators, orders, and shared equipment interact, while authorization from the actor entitled to permit conduct in the destination space to the robot performing an individual task is less often made explicit. This paper compares public documents and technical specifications concerning mixed-use urban facilities, access-controlled apartment buildings, and drone delivery. It identifies settings in which authority to control infrastructure, authority to permit entry into a destination space, and authority to permit sensing or recording do not coincide. The paper then states a Non-Inference Principle at Authority Boundaries and identifies the minimum information and decision criteria needed to reflect the allocation of authority across spaces in an automated access decision. It does not propose a new policy language or authorization protocol. Rather, robot access is used to examine, as a concrete design problem, who may permit what at a spatial boundary.