The ethical alternative to griefbots
A griefbot is an AI system that simulates a dead person, usually built after they have died from the messages, recordings, and posts they left behind, so that the bereaved can carry on a kind of conversation with them. The ethical alternative to a griefbot is not to abandon the wish behind it, which is deeply human, but to invert the order of consent: instead of recreating someone after death without their agreement, you build a governed representation of yourself while you are alive, having chosen it freely. That single difference, who consented and when, separates a practice many ethicists warn against from one a person can stand behind.
This page sets out what a griefbot actually is, where the serious ethical problem lies, what researchers and ethicists have said about it, and what a consent-first alternative looks like in practice. The aim is to be fair rather than alarmist. The impulse to keep talking to someone you have lost is not a flaw to be scolded. The question is only how it is done, and whether the person being recreated ever had a say.
The core ethical problem: recreating someone who never consented
The defining feature of most griefbots, and the root of the ethical difficulty, is that they are built of the dead, not by them. A grieving family or a company gathers up a person's old texts, voice notes, and social posts and trains a model to speak as them. The deceased is the raw material. They are never the author, and crucially they were never asked.
This matters because a convincing simulation makes claims in a person's name. It produces sentences they never said, opinions they may never have held, reassurances they never offered. The dead cannot correct it, cannot object, and cannot withdraw. A representation that the subject did not consent to and cannot govern is, at bottom, someone speaking for a person who can no longer speak for themselves, and presenting it as that person's own voice. Whatever comfort it offers, it begins from a place the subject never agreed to stand.
A griefbot is built of the dead. A Persona is built by the living.
The consent line
Almost every meaningful ethical question about this technology resolves to one line: was the person represented a consenting author, or an unconsenting subject? On one side is the typical griefbot, assembled after death from data the person left for other purposes, governed by whoever holds that data, answerable to no one. On the other is a representation a person built themselves, while alive, deciding what it contains and what it will never claim.
This is the distinction between a griefbot or a deadbot, terms for the after-the-fact recreation, and a consent-first Persona authored in advance. The words sound similar and the technology overlaps, but ethically they are near opposites. One takes a voice; the other is given one. The presence or absence of the subject's own consent is not a detail. It is the whole question.
Consent here also has to mean more than a checkbox ticked once. It means the person chose what went in, can shape how it behaves, and set the rules for who may reach it and when. Consent that the subject cannot exercise because they are already gone is not really their consent at all. That is why the timing matters as much as the agreement: the only person who can truly consent to being represented is the living one.
What the research and ethicists say
Serious commentary on this technology has grown more cautious as the tools have improved. Coverage in outlets such as Scientific American has weighed whether griefbots genuinely help mourners or risk holding them in place, and has noted that the evidence is far from settled. The careful conclusion across much of this writing is not that the technology is worthless, but that its benefits are conditional and its risks are real.
Academic and ethical discussion, including work surfaced in venues like The Conversation, has pressed harder on the consent question specifically. Researchers in this area have argued that recreating a person without their prior agreement can violate their dignity, that families and companies can have interests that diverge from what the dead person would have wanted, and that there should be guardrails preventing a simulated person from being used, monetised, or made to say things the original never would. The recurring theme is governance: who controls the representation, on whose authority, and with what limits.
Read together, this body of work does not say the wish to stay connected is wrong. It says that wish should be served in a way that respects the person being represented, and that consent and control are the conditions under which it can be. That is a standard a consent-first approach is designed to meet, and one an after-the-fact griefbot structurally cannot.
The documented harms of non-consensual deadbots
Beyond the question of principle, several concrete harms have been documented or seriously argued. They are worth naming plainly, because they are the reasons caution is warranted, not abstractions.
Recreation without consent: the most basic harm is representing a person who never agreed to it, putting words and views in their mouth that they cannot contest.
Interference with grief: a simulation that is always available can, for some people, delay the work of mourning rather than ease it, keeping the bereaved tethered to a presence that cannot truly change or grow.
Commercial capture: where a griefbot is run by a company, the dead person can become a product, with their likeness retained, monetised, or repurposed by a party they never authorised.
Drift and fabrication: a model trained to sound like someone will fill gaps by inventing, producing statements the person never made and the family may wrongly take as authentic.
None of these harms is inevitable in every case, and that is precisely the point. They follow from a structure, building the representation after death, without consent, under outside control, that almost guarantees at least some of them. Change the structure and most of these risks fall away. Tools designed for the bereaved to talk to a dead loved one through AI or to talk to an AI version of someone who died sit on exactly this fault line, and where they are built without the subject's consent, they inherit exactly these problems.
The consent-first alternative: build your own while alive
The alternative is straightforward to state. Rather than letting someone be recreated after death without their say, a person builds their own representation while alive, as the author of it. At Afterlife AI™, that is a Persona: a governed, consent-first record of who you are, drawn only from what you actually provide, across the many dimensions of a real person rather than scraped from leftover data.
Because the subject is the author, the harms above are addressed at the root rather than patched afterward. There is consent, because you chose to make it. There is no fabrication of the kind griefbots risk, because it draws only on verified memory you supplied, rather than guessing to fill silence. And there is governance, because you set the terms while you can. The wish to stay connected is honoured, but the person at the centre of it kept their voice in the deciding.
The only person who can consent to being represented is the living one.
How to do it responsibly
Doing this well is less about the technology than about the conditions around it. A representation of a person should be authored by that person, drawn from verified memory rather than inference, and placed under clear, lasting control over who may reach it and when. Those conditions are what separate a respectful practice from an extractive one, whatever it is called.
Control is the part that is easiest to overlook and most important to get right. The Executor Lock™ is built to provide it: it governs who may activate a Persona and when, gives a named executor the final word, and makes the result permanent once set, so it cannot be retrained, altered, or commercialised after your death. That permanence is what turns a representation from something that could be taken and changed into something that stays as the person left it. It is the structural answer to the commercial-capture and drift harms that worry ethicists most.
If what you want is a place for those you love to return to, the same principles produce a governed AI memorial that respects the person it remembers, because the person built and bounded it themselves. The difference between this and a griefbot is not the comfort it offers the bereaved, which can be real in both. The difference is that here, the comfort does not come at the expense of the consent of the one being remembered.
So the ethical alternative to a griefbot is not refusal of the wish behind it. It is a reordering of it. Recreating the dead without their agreement asks the bereaved to accept a voice the subject never approved. Building a Persona while alive, governed and consent-first, lets a person decide for themselves how they will be present after they are gone, and keeps that decision protected. The technology is similar. The ethics are not. The whole of the difference is who consented, and when.