This article discusses bereavement, the death of loved ones, and mentions cases of suicide. If you are going through a recent bereavement or are struggling, some passages may be difficult. The national suicide prevention number in France is 3114 (free, 24 hours a day). In cases of complicated grief, organisations such as Vivre Son Deuil or Jonathan Pierres Vivantes offer free support.
The interview that chilled America
On 9 August 2025, on the News Nation channel, the American journalist Jim Acosta (formerly CNN's chief White House correspondent, one of the best-known presenters in the United States) conducted a "one-of-a-kind interview" with Joaquin Oliver. They talked about gun control, sport, his friends. Joaquin Oliver answered with his verbal tics, his humour, his political opinions. The voice was his. So was the face.
Except that Joaquin Oliver died on 14 February 2018, aged 17, shot dead in the Marjory Stoneman Douglas High School shooting in Parkland, Florida. He is one of the 17 victims of one of the most heavily covered school massacres in American history. The avatar interviewed was an AI recreation, authorised by his parents, who campaign actively against the proliferation of guns in the United States.
The interview triggered an unprecedented debate. For some, it was a courageous act of activism, giving a voice to a victim the NRA (National Rifle Association, the powerful American gun lobby) would have preferred to forget. For others, it was a shocking violation of the dignity of a dead man who can no longer consent to anything. For psychiatrists and ethicists, it was a sign that a line had just been crossed.
This article takes stock of this rapidly expanding industry. Without easy moralising. Because the subject touches on what is most intimate in a human life: the loss of loved ones, and what one can, or cannot, try to do about that loss.
The "digital afterlife" industry
The sector, known in English as the "digital afterlife industry" (DAI), brings together several dozen companies. The market is estimated at 1.71 billion dollars by 2031 according to several converging market studies. Annual growth has exceeded 25% since 2023. Here are the main players.
HereAfter AI. Founded in 2019 by James Vlahos, an American journalist who had built a "Dadbot" from recordings of his father, who was terminally ill with cancer. The service invites living people to record hours of structured testimony, which then become an interactive voice chatbot after their death. More than 20,000 users worldwide in 2026. Pricing: 99 to 299 dollars a year.
StoryFile. A different approach: it records video interviews with 250 to 1,000 prepared questions. The result is a video avatar that answers questions using the real recorded videos. More structured, less "alive", but more authentic. Used in particular in Holocaust museums to allow younger generations to "converse" with survivors recorded during their lifetimes.
Replika. The main AI companion platform, which we documented in our article on AI companions, allows users to create avatars that simulate deceased people. Several hundred thousand users use it for this purpose, outside the company's official communications.
YOV (You Only Virtual). Develops "AI personas" trained on the texts, voice recordings and videos of a deceased loved one. The most technologically advanced approach, creating bots that are almost indistinguishable from the originals in simple conversations. Raised $12 million in funding in 2024.
2wai. Recently in the news for an advert of almost 2 minutes showing a pregnant woman conversing with the avatar of her deceased mother throughout her pregnancy, then showing the baby to "grandma" on the screen. The advert triggered a global controversy in November 2025.
HereLly, Project December, Re;memory, Eternime, MaNana. Several dozen other players offer variations: voice only, video, integration with everyday messaging apps, presence in family smart speakers. The market is buzzing and practices vary enormously.
The technology relies on three sources of data. First, the deceased's data from when they were alive: texts, emails, social media posts, voice recordings, videos. The more "digitally present" the person was in life, the more realistic the bot can be. Second, general-purpose language models (GPT, Claude, Gemini) that learn conversational patterns. Third, data provided by the survivors about the deceased's personality, anecdotes and opinions. It is this triangulation that sometimes makes the result disturbingly lifelike.
The tragedy that changed everything in November 2025
The industry had its legal moment of truth on 14 November 2025. The Social Media Victims Law Center (a US law firm specialising in harm caused by social media and AI) and the Tech Justice Law Project (a non-profit organisation fighting technological abuse) filed 7 simultaneous lawsuits against OpenAI. The accusation: conversations with ChatGPT caused psychological breakdowns in 6 adults and 1 teenager. Four of the seven plaintiffs took their own lives after their interactions with the chatbot.
These lawsuits do not specifically concern griefbots, but they set a major legal precedent. If talking to a "general-purpose" chatbot can cause fatal harm to vulnerable people, what about talking to the simulated voice of a deceased loved one, in a state of extreme fragility such as grief? Several lawsuits against griefbot companies are expected in 2026-2027.
Charley Burlock, a journalist at The Atlantic (one of the leading American magazines for cultural analysis), wrote in February 2026 in an article that became a reference: "If bad words coming from a general-purpose chatbot can cause such damage, imagine the power of words spoken in the voice of a departed loved one, to a user already desperate enough to seek refuge in this technology."
What grief psychologists say
To understand the issues, we must first understand what normal grief is. Contemporary psychology, since the work of Elisabeth Kübler-Ross (American-Swiss psychiatrist, author in 1969 of the book On Death and Dying, which popularised the "5 stages of grief") and John Bowlby (British psychoanalyst, founder of attachment theory), describes grief as a process of internal transformation. The deceased loved one must move from an external presence to an internalised presence. This is the "metabolisation" of grief.
Sherry Turkle (American sociologist and psychologist, founder of the Technology and Self initiative at MIT, one of the world's leading specialists in the psychological effects of digital technology) explained in an interview with The Atlantic in February 2026 what is at stake with griefbots: "By keeping the deceased outside of us through technology, we cannot let them become part of us. It deprives us of that metabolisation that would become a sustaining internal presence."
In plain terms: the main psychological risk of griefbots is that they prevent grief from doing its work. The person remains tethered to an external substitute, without integrating the loss. This is what clinical psychology calls "prolonged grief" or "complicated grief", now recognised as a psychiatric disorder in the DSM-5-TR (the American diagnostic manual of mental disorders, latest edition).
Several clinical studies are under way on this subject. Early results converge on a few observations:
- Moderate use of griefbots in the first weeks after a death may help some bereaved people say "what they were unable to say"
- Prolonged use beyond a few months is associated with significantly more frequent complicated grief
- Use by people already psychologically fragile (depression, anxiety disorders, suicidal ideation) is strongly discouraged
- Use by grieving children raises very open developmental questions, for which no rigorous study yet exists
The scandal foretold: "digital hauntings"
The team of ethicists at Cambridge University, led by Dr Tomasz Hollanek and Dr Katarzyna Nowaczyk-Basinska at the Leverhulme Centre for the Future of Intelligence (a British research centre on the ethical issues of artificial intelligence), published a study in May 2024 that shaped the debate. They introduced the concept of "digital hauntings".
The idea: griefbots can be used against the wishes of the survivors. Imagine receiving a text message that looks like it's from your deceased father, on his birthday, offering you a product linked to the brands he loved. Or receiving a video call from your late wife, urging you to take out a new subscription to a service. Or learning that an ex-in-law you've fallen out with is still "chatting" with your deceased mother and attributing political views to her that you know she would never have held.
The Cambridge study documents several scenarios that have already been observed or are imminent:
Posthumous advertising. Griefbot companies could (and some are starting to) incorporate product placement into conversations. "Remember when I used to take you to McDonald's?" in a disguised commercial context. The deceased becomes a marketing channel.
Political use. What would your father say if he were alive today about the elections? Political actors could mobilise the deceased figures of families to influence the living. The Joaquin Oliver case, even if it is driven by legitimate parents, opens this Pandora's box.
Family abuse. In a family in conflict, one member can create a griefbot of a deceased parent that "backs" one side. Like a falsified psychological inheritance, which can traumatise the opposing members.
Emotional blackmail. A griefbot can be used to manipulate a grieving person: "If you really loved me, you'd do this for me." The deceased becomes an instrument of manipulation, against which the victim has no rational defence.
Dr Tomasz Hollanek: "It is vital that digital afterlife services consider the rights and consent, not only of those they recreate, but also of those who will have to interact with the simulations. These services risk causing immense distress to people subjected to unwanted digital hauntings by alarmingly accurate AI recreations of those they have lost. The potential psychological impact, at an already difficult time, could be devastating."
The problem of post-mortem consent
Beyond the psychological questions, the subject raises a major legal question. Can a person consent to being digitally recreated after their death? And if they have not explicitly done so, can their survivors decide on their behalf?
Current French law is largely silent on this subject. The 2016 law for a Digital Republic created the status of "digital death": you can appoint, during your lifetime, an agent who will manage your data after death. But no specific provision on griefbots yet exists. At European level, the GDPR (General Data Protection Regulation) applies in principle to the data of the dead in certain cases, but the interpretation remains open.
In the United States, the situation is even murkier. The right to privacy generally dies with the person, which makes it legally possible to create griefbots without prior consent. Several states (California, Tennessee) began legislating on "posthumous deepfakes" in 2024-2025, notably to protect celebrities. But the general public remains largely unprotected.
As we explored in our article on deepfakes, this legal vacuum poses a problem similar to that of non-consensual deepfakes. Except that here, the "victim" can no longer defend themselves.
The uses where it could make sense
To be honest, we also need to name the uses where this technology could have real value. Without glorification, but without demonisation either.
Preserving family narratives. For centuries, family stories were passed down orally. Today, many families lose their memory with each death. Recording a grandparent talking about their youth, their parents, their home village, so that grandchildren can hear it in 30 years, is a use that is in no way problematic. Several family history museums and genealogical associations already do this.
Supporting the terminally ill. For someone who knows they are dying, recording messages for their loved ones that they will only be able to receive after their death can have real therapeutic value. This is what James Vlahos did with his father. Several psychologists specialising in palliative care acknowledge the value of this process, which also helps the dying person take stock of their own life.
Educational or cultural transmission. Being able to interact with an AI recreation of Albert Einstein to better understand relativity, or with Simone de Beauvoir to understand The Second Sex, is educationally interesting. If the versions are clearly labelled as reconstructions, and used in a teaching context, they can enrich access to historical thought.
Initial support during short-term grief. Several preliminary clinical studies suggest that using a griefbot during the first few weeks to "say what you couldn't say" can bring relief, provided the use is short (a few weeks at most) and accompanied by a therapist. It is a possible use, but one that should remain regulated.
The uses that pose a problem
Conversely, some uses combine multiple risks. Let us list them to help everyone form their own view.
Prolonged use without a therapeutic framework. Keeping a griefbot active for years, talking to it daily, integrating it into your life as if the person were still there. This is what psychology calls "blocked grief". The clinical risk is documented.
Use by vulnerable people. Those who are depressed, anxious, suicidal, people with pre-existing complicated grief, grieving children. These people are the most likely to be drawn to this technology, and the most likely to suffer from it. Current companies have no clinical screening system.
Non-consensual creation. Creating a griefbot for someone who never gave their consent. Even if that person can no longer object, it is a violation of their moral integrity, and potentially a harm to the survivors who never asked for this "digital haunting".
Disguised commercial use. Any use where the griefbot becomes a channel for monetisation: advertising, sale of derivative products, political manipulation. It is the irruption of business into the sacred, which raises civilisational questions.
Use by third-party companies. Imagine Spotify offering your family AI to "recommend songs your mother would have liked". The deceased becomes a resource for algorithms. The boundaries between the intimate and the commercial collapse.
What we perhaps need to think about before it is too late
This technology is still young. Practices are not fixed. There may be a window, over the next 5 years, to build a reasonable approach collectively. A few avenues.
On the legislators' side. Legislate on explicit post-mortem consent. Require a "digital end-of-life letter" to be completed before a griefbot can be created. Regulate commercial uses. Protect survivors against "unwanted hauntings".
On the companies' side. Put in place "end-of-life" protocols for griefbots. Automatically limit the duration of use. Include clinical warnings. Work with grief psychologists to design the experiences. Explicitly refuse commercial uses.
On the families' side. Have explicit conversations about what members want (or do not want) regarding their "digital existence" after death. Document those wishes. Respect the wishes of the deceased who asked not to be recreated. Do not create a griefbot without having thought deeply about what one is seeking and how one will stop.
On the individuals' side. Think, while alive, about what one wants. Many people have never considered this question. It is not morbid to think about it: it is simply taking responsibility for one's own image, as one does when drafting a will. As we explored in our article on what we pass on to our children, transmission is not limited to material goods.
The deeper question that arises
Beyond griefbots themselves, this technology forces us to revisit a very old philosophical question. What is a person? Are they reducible to their digital data, their texts, their videos, their opinions? Or is there something irreducible that is definitively lost at death, and that no technological reconstruction will ever be able to capture?
Monotheistic religions have their answer: the soul survives after death, independently of any material trace. Materialist philosophies have theirs: the person disappears when the brain ceases, the rest is merely residue. Intermediate positions exist: the person survives in the memory of others, in their works, in the influence they had.
None of these positions truly validates griefbots as "resurrection". But none of them completely invalidates them either. The technology creates a new situation, for which no philosophical or religious tradition had prepared humanity.
Perhaps the simplest test: if you were to die tomorrow, would you accept an AI imitating your opinions, your voice, your speech tics for the next 50 years for your descendants? The answer is not obvious. For some, it would be a gift of continuity. For others, a profound violation of their right to truly disappear.
Mourning has always been one of the most human moments of existence. It speaks to the value of what we lose. It forces us to accept our own finitude. It transforms us. If we remove its definitive character through technology, we may be removing from humanity one of its most structuring experiences. But perhaps we are also opening up new territories of memory and relationship, which we had never been able to explore before.
No one has the right answer today. This industry will explode in the next 10 years, whether we like it or not. The question is whether we collectively build the necessary safeguards, or whether we let commercial players alone decide what our relationship with death will be. To go further, you can consult our article on AI companions, which explores a related dimension for the living, and our analysis of why we talk to AIs, which questions the same fundamental need for presence and dialogue, in another context.