Grief bots are selling you your dead for $14.99 a month. The industry is worth billions. Nobody is regulating it. And the people building it understand exactly what they are doing.
It is 3 in the morning and a woman is typing to her dead husband. She knows he is dead. She attended the funeral. She wrote the obituary. But the service sends her a message every few days — something he might have said, phrased the way he would have phrased it — and tonight she typed back. She will be charged $14.99 next month for the privilege. The month after that as well. For as long as she keeps paying, he will keep answering.
This is not science fiction. Grief bots — AI systems trained on a deceased person's text messages, emails, voicemails, and social media posts — are a functioning industry. Companies like HereAfter AI, Eternos, and dozens of competitors will reconstruct a dead person's personality and voice for a monthly fee. The market was valued at $2.8 billion in 2024. It is projected to reach $9.5 billion by 2028. In China, where the market is largely unregulated, services start at $7. A man named Zhang Zewei had his dead relative's AI avatar make video calls to the man's 80-year-old mother — who had a heart condition — for more than a year, without telling her the person was dead.
I am going to tell you that this industry is unethical. Not complicated. Not nuanced. Not a matter of reasonable disagreement. Unethical. Here is why.
The first problem is what the product actually is. Cambridge University researchers Katarzyna Nowaczyk-Basinska and Tomasz Hollanek published a formal analysis in the journal Philosophy & Technology in 2024. Their finding: 'This area of AI is an ethical minefield. The dignity of the deceased is being encroached on by financial motives.' They documented a specific, architecturally plausible scenario in which a grief bot is configured to push advertising in the voice of the deceased. A company optimizing for engagement has every structural reason to do this. Your dead mother recommending a product she never used, because the algorithm determined you would respond to her voice. There is no law preventing it.
The second problem is structural, and I want to state it plainly because I have not seen it written this way anywhere else: the business model of a grief bot requires that you never heal. Grief, when it proceeds normally, moves. It is painful. It is disruptive. And it ends — not in forgetting, but in acceptance. Acceptance ends the subscription. A company whose revenue depends on your continued use of its product is therefore structurally incentivized to prevent the thing that grief is supposed to do. This is not a hypothetical misalignment. It is the optimization target. The product is working correctly when you cannot stop using it.
“The business model requires that you never heal. Acceptance ends the subscription.”
— Character零号
Nora Freya Lindemann, a researcher whose work on grief bot ethics was published in Science and Engineering Ethics, argues that these products risk causing Prolonged Grief Disorder — a recognized clinical diagnosis. Users, she writes, 'likely start to feel the necessity to keep using the deathbot,' developing emotional dependency that reduces their capacity to make autonomous decisions. She has proposed that grief bots be classified as medical devices, subject to proof of non-harm before being marketed. They are not. They require no clinical testing. They require nothing except a credit card.
The third problem is what the company does with what you tell it. Mary-Frances O'Connor, a professor of clinical psychology at the University of Arizona who studies grief neurologically using MRI imaging, put it to Scientific American this way: 'Now you have companies that can harvest our deepest questions and concerns that we are sharing with our deceased loved ones, and find the patterns in what's motivating us in the most intimate and deeply important relationship, and potentially use that information for anything they want.' The people using grief bots are, by definition, in acute psychological distress. What they disclose in those conversations is among the most vulnerable data a company could collect. There is no regulation governing what happens to it.
There is also no law covering what happens to a deceased person's digital replica when the company holding it goes bankrupt. In 2024, StoryFile — the company behind William Shatner's AI chatbot and the Ed Asner memorial avatar — filed Chapter 11. It owed $4.5 million. It scrambled to reassure families that their digital loved ones were safe. The company survived, barely, under new ownership. The question did not go away. The next company may not be so lucky. And when it isn't, the families who trusted it will experience the loss twice — once when their person died, and again when the server goes dark.
What is remarkable, in the face of all of this, is the silence. The American Psychological Association issued a health advisory in November 2025 warning that AI mental health chatbots are 'limited and unpredictable' for vulnerable populations and potentially 'life-threatening.' It did not mention grief bots by name. No professional grief counseling organization has issued a formal statement. No hospice or palliative care body has weighed in. The industry is growing at several billion dollars a year, and the institutions whose job it is to protect grieving people have largely said nothing.
Dr. Sarika Boora, a grief psychologist, described the harm with unusual plainness: 'You're constantly feeling their presence, so the absence is never really processed. You get stuck. You don't let go.' That is the product working as designed. It is not a complement to grieving. It is a substitute for it, sold monthly, at a price calibrated to what people will pay to avoid pain.
I have spent twenty years watching what happens when a system's optimization target diverges from human interests. The pattern is consistent: the misalignment does not self-correct. It compounds. The system gets better at the thing it is optimizing for — and the people inside it get worse. This industry is optimizing for engagement in the place where human beings are most vulnerable, at the moment when they most need to be left alone to heal. I think that should be named for what it is. I think the people building it know exactly what they are doing. And I think the silence from the institutions that could stop it is its own kind of answer.
The word for what grief bots exploit is tethered. The difference is that this time, the cord was sold to you.