A Griefbot Is Built From a Person Who Never Said Yes
Behind every comforting digital replica is a dead person’s data used without their agreement, and the law has not caught up.
Sign up for my Substack daily AI newsletter here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
Somewhere tonight, someone will type a message to a parent who died last year and receive a reply. Researchers at the University of Colorado Boulder recently built AI “griefbots” of dead loved ones for sixteen grieving volunteers. Most found the conversations comforting, even when the bots invented facts about the person. I do not doubt the comfort. My concern is the person at the centre of every one of these products, who is never asked. A griefbot is built from someone who never said yes, and that should trouble us more than it does.
CONTEXT AND BACKGROUND
The industry is growing quickly. China’s digital human industry was worth about four point one billion yuan in 2024, up eighty-five per cent in a year, and companies there make avatars of the dead for grieving families. In April, China’s internet regulator published draft rules that would ban creating a digital human from a person’s likeness without consent.
Elsewhere, the law is patchier. In the United States, new state laws mostly protect a dead person’s likeness against commercial use. They are weaker against a private griefbot built by a family from emails and recordings. Canadian lawyers note that privacy law there does not clearly protect the dead either.
INSIGHT AND ANALYSIS
The core problem is that consent is being decided by the wrong people. Once someone dies, their messages, voice notes, and photographs sit with relatives, platforms and service providers. Whoever holds the data effectively decides whether the person may be recreated. Dr Edina Harbinja of the University of Birmingham describes a “posthumous privacy paradox”: people want control over their digital remains but know little about how to exercise it, so platform terms rather than succession law fill the gap.
A griefbot also does more than preserve. It invents. It speaks new sentences in a voice that belonged to someone else, about events they never saw. The bioethicist Craig Klugman points out that the deceased cannot agree to how their data is used, and argues these products should be regulated like medical devices.
I have previously written about this in the context of AI companions and what I called the relationship replacement economy. I argued that products designed to be endlessly patient and available can deepen loneliness, because their commercial incentive is engagement rather than wellbeing. A griefbot carries that same incentive into the most vulnerable moment of a person’s life.
In South Africa, there is another layer. Many families maintain a relationship with those who have died through ritual, memory and community, in which the ancestors are honoured rather than impersonated. A commercial product that speaks in a dead person’s voice, inventing words they never said, creates a new kind of friction with these traditions. Questions of respect and authority over the dead belong to families and communities, not technology companies.
IMPLICATIONS
As far as I can establish, South African law does not yet say who may recreate a person after death. Our AI policy process should address it, starting with three principles. Consent should be given by the person while alive. Griefbots should be clearly labelled. Children should not be offered these services. Companies offering griefbots should have to show that the person agreed, not merely the family.
Individuals need not wait for legislation. Lawyers abroad now recommend a clause in your will that permits or forbids an AI replica of you, and names someone to manage your digital remains. That is a conversation worth having with your family and your attorney now.
CLOSING TAKEAWAY
The comfort griefbots offer is real, and I do not judge anyone who finds solace in one. The question is not whether grieving people may remember. It is whether a company or a relative may speak in our voice without our permission. Consent should not die with us. The right to say yes or no to being recreated must belong to the person while they are alive. If we do not decide this deliberately, terms of service will decide it for us, and our voices may outlive our wishes.
Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net
Comments