Masaki Iwasaki, Human-Like Epistemic Trust Without Human-Like Legal Accountability?Conversational AI in Mental Healthcare(2026)
<Abstract>
This paper examines conversational AI (CAI) in mental healthcare through the lens of epistemic trust and legal accountability. Building on Sedlakova et al.’s framework of “human-like epistemic trust,” it argues that although CAI cannot satisfy the normative conditions underlying therapeutic trust or bear moral responsibility itself, developers and providers who design systems to cultivate such trust should bear accountability for the foreseeable reliance these systems induce.
The paper identifies a structural mismatch in current CAI governance: systems are increasingly designed to appear empathic and therapeutically competent, while accountability for harmful outcomes remains unclear or diluted through disclaimers and regulatory gray zones. Drawing on the seven conditions of epistemic trust, the paper proposes a responsibility-sensitive framework for mental-health CAI governance, linking therapist-like design features to duties of transparency, risk management, human oversight, and clinically grounded safeguards. It further develops the “fictional-character” account of CAI, arguing that because conversational AI functions as a designed persona rather than a genuine moral agent, accountability must ultimately rest with the human actors who create and deploy these systems.
Keywords: Conversational AI (CAI), Mental Healthcare, Epistemic Trust, Legal Accountability, AI Governance, Therapeutic Chatbots, Responsibility-Sensitive Design, Digital Mental Health, AI Ethics, Fictional-Character Account
Masaki Iwasaki, Human-Like Epistemic Trust Without Human-Like Legal Accountability?Conversational AI in Mental Healthcare(2026)
<Abstract>
This paper examines conversational AI (CAI) in mental healthcare through the lens of epistemic trust and legal accountability. Building on Sedlakova et al.’s framework of “human-like epistemic trust,” it argues that although CAI cannot satisfy the normative conditions underlying therapeutic trust or bear moral responsibility itself, developers and providers who design systems to cultivate such trust should bear accountability for the foreseeable reliance these systems induce.
The paper identifies a structural mismatch in current CAI governance: systems are increasingly designed to appear empathic and therapeutically competent, while accountability for harmful outcomes remains unclear or diluted through disclaimers and regulatory gray zones. Drawing on the seven conditions of epistemic trust, the paper proposes a responsibility-sensitive framework for mental-health CAI governance, linking therapist-like design features to duties of transparency, risk management, human oversight, and clinically grounded safeguards. It further develops the “fictional-character” account of CAI, arguing that because conversational AI functions as a designed persona rather than a genuine moral agent, accountability must ultimately rest with the human actors who create and deploy these systems.
Keywords: Conversational AI (CAI), Mental Healthcare, Epistemic Trust, Legal Accountability, AI Governance, Therapeutic Chatbots, Responsibility-Sensitive Design, Digital Mental Health, AI Ethics, Fictional-Character Account