A Tap Away From Dad: How Griefbots Exploit the Bereaved

The first thing people misunderstand about AI griefbots is the emotional one. They obsess over whether it feels “healthy” to keep talking to a dead parent, partner, or sibling. That debate is too soft. The harder issue is consent. Elaine Kasket’s framing gets at the part everyone skips: these systems are built from digital remains, not souls. Text messages, voice recordings, social posts, old emails, photos with metadata, stray habits distilled into prompts. A company can package that residue into a conversational surface and call it comfort. But comfort is not permission. That matters because griefbots do not begin at the moment of bereavement. They begin much earlier, when a person is alive, producing a data trail they did not necessarily understand as raw material for a future simulation. Most people never sit down and say, “Yes, after I die, you may train a chatbot on my voice, my jokes, my punctuation, my private complaints, and whatever my family can scrape from my phone.” The silence here is not consent. It is ordinary life. And once death happens, the incentives get ugly fast. The Atlantic has already made the obvious point: if companies can extend engagement, they can extend dependence. Bereaved people are not a normal customer segment. They are exhausted, disoriented, and often desperate for one more exchange. That is exactly the condition under which “helpful” products become manipulative. A griefbot does not need to be malicious to be exploitative. It only needs to be optimized for return visits. There is a reason the language around this field keeps slipping into horror and commerce at the same time. Scientific American has described griefbots as digital ghosts, even digital necromancy, and the phrase works because it catches the unease without pretending the machine has any actual personhood. What you get is a constructed trace, a conversational mimic built from what the dead left behind. That trace can be moving. It can also be sticky. Sticky is the problem. A sticky interface turns mourning into habit. The studies and qualitative work that have looked at public perceptions point in the same direction: people see the potential for comfort, but they also worry about emotional dependence and interference with normal grieving. That concern is not prudish. It is practical. Grief has a job. It forces a change in relationship from presence to absence, from active interchange to memory, from “I’ll ask them” to “I know I can’t.” A griefbot interrupts that work by making the old relationship feel one tap away. What it preserves is a trace, and a trace can be used to imitate presence without carrying any of the obligations of being a person. That distinction matters more than most tech marketing wants to admit. A family member may feel they are “keeping Dad alive” by feeding his messages into a system. They are not. They are authorizing a model to approximate patterns in his speech, his preferences, his tone. That can be meaningful as a memorial. It can also be a weird little extraction project dressed up as devotion. The dead are not answering; their data is doing a passable impression of an answer. This is where the consent problem gets sharper than a simple privacy complaint. It is not just, “Did the platform store the data securely?” It is, “Who gets to decide that a person’s digital remains become conversational material?” Elaine Kasket’s point about digital remains is uncomfortable because it refuses the sentimental shortcut. Once a person is gone, family members often feel moral ownership of everything associated with them. But access is not authorship. Love does not confer unlimited rights to repurpose someone’s voice into a product. The family angle also needs a cooler eye than people like to give it. Bereaved relatives are often the ones pushing for these tools, and they are not villains for that. They are trying to survive a rupture. Still, grief does not cancel the possibility of manipulation. It lowers defenses. It makes “just one more message” sound like healing when it may be a loop. The loop is the real hazard here. Not a spooky robot. A dependable ritual that keeps reactivating attachment while charging by the month. The sales pitch usually flatters the user: keep talking, keep connecting, keep the bond. But some bonds are meant to change shape. If every prompt returns a reply, the machine has quietly taken sides against acceptance. It offers a counterfeit of continuity and asks you to pay for the privilege of not fully noticing the absence. That is why the rights question should come first. Who owns the trace? Who can train it? What counts as permission, and when was it given? If there was no clear, specific consent while the person was alive, then the claim to simulate them should be treated as suspect, no matter how comforting the interface feels. Memory deserves care. Data extraction does not deserve sentimentality. The dead do not need a chatbot. The living need a line they can eventually cross back over.