Words They Never Said: Generative Griefbots, Posthumous Attribution, and an Islamic Ethics of Speaking for the Dead

Abstract

Chatbots built from the messages, voice recordings and posts of people who have died are now sold as aids to mourning. Most ethical discussion of these "griefbots" asks whether they help or harm the bereaved, or whether the deceased consented to the use of their data. This article argues that generative griefbots raise a further and more basic problem, which it calls the attribution problem. A system built on a large language model does not replay what a person said. It produces new sentences, in that person's voice, that the person never said. Each output is a fresh attribution of speech to someone who can no longer confirm, correct or withdraw it. The article develops this problem through the literature on posthumous harm and tests it against the fictionalist reply that users know the bot is not the person. It then turns to a resource that the field has largely ignored: Islamic ethical and juristic thought about speaking for the dead. A 2025 ruling by Egypt's Dar al-Ifta permits AI animation of the deceased for remembrance but forbids making them say what they never said. Read alongside the hadith tradition's long discipline of verified attribution, this ruling points to a practical standard. The article proposes "attribution-bounded" design: quotation before generation, visible provenance for every reply, a ban on new commitments in the dead person's name, and mourning-sensitive limits on use. It closes with implications for regulation in the Gulf and beyond.