<p>This article examines artificial empathy under conditions of AI opacity. It argues that the ethical problem is not simply whether AI really feels, but how human-AI relations should be governed when systems produce empathy-like responses whose inner status remains uncertain. XAI can make artificial empathy more inspectable and can support debugging, contestation, and accountability, but it cannot by itself determine whether outwardly caring behavior corresponds to inner affective life. To clarify the problem, the article distinguishes functional responsiveness, social uptake, and phenomenological feeling. It then develops a Zhuangzian model of non-mastery as a normative framework for responsible action under partial understanding. “Dao tong wei yi” (道通為一) supports relational accountability by locating AI effects within relations of design, data, deployment, reliance, benefit, and harm. “Xinzhai” (心齋) supports suspended attribution by denying that empathy-like performance alone entitles users or institutions to attribute inner affective life to AI. Cook Ding’s skill supports institutional joints: the points at which opaque systems can be challenged, corrected, overridden, monitored, or prohibited. The article contributes to AI ethics by showing how responsible human-AI relations can be governed without requiring total transparency or resigning before opacity.</p>