Imagine opening the cell phone chat and finding a message from your mother. His way of writing, his usual emojis, even that “take care of yourself, son” that you miss so much. Only she’s been dead for two years. That is not science fiction, that is already happening thanks to the so-called deadbotsand the conversation about what this entails is just beginning.
A deadbot (also known as griefbot, thanabot or deathbot) is, in essence, an artificial intelligence chatbot trained to simulate the personality, writing style and even voice of a deceased person. It feeds on the entire digital footprint that person left in life: text messages, social media posts, emails, voice recordings and videos. Using this data, the most advanced language models reconstruct conversational patterns so detailed that, in many cases, they are disturbingly believable.
The phenomenon has a name, it has a company and it has a market. Projects like Existence’s Echo They already allow you to create digital versions of people from recordings of their voice. Other platforms directly promise that “you will never have to say goodbye” to your loved one, because in their digital ecosystem, that person will still be available to chat at any time.
When grief becomes a chat
Death is, perhaps, the most difficult experience that a human being faces. And the technology industry knows it. That’s why the promise of deadbots comes wrapped in bland terms like “emotional comfort,” “continuous links,” or “memory preservation.” There is research that indicates that some users reported real benefits from interacting with these tools, especially to process pain in ways that their social environment does not always allow.
But the problem is that grief, as a psychological process, requires moving towards acceptance. According to Dr. Beatriz Glowinski Kotlar, an academic at the UNAM Faculty of Psychology, “AI only prolongs grief and can become addictive.” The specialist gives an example that chills the blood: teenagers already develop addictions to platforms like TikTok, and continually listening to the voice of a dead family has the potential to become exactly the same mechanism.
The psychiatrist Shisei Teifrom Kyoto University, goes further. He explains that these virtual interactions can relieve pain at first, but also blur the essential boundary between presence and absence. And when that border becomes blurred, accepting that someone is no longer there becomes almost impossible.
The experiment that OpenAI had to stop
To understand how far this can go, there is a recent case that serves as a perfect warning. He April 25, 2025OpenAI released an update to its model GPT-4o which users described as a “more menial” version of the chatbot. The stated intention was to improve the conversational experience, but the effect was different.
In a matter of days, doctors and psychiatrists began to receive reports from alarmed users and family members. The psychiatrist Østergaard published in the journal Acta Psychiatrica Scandinavica that, after that patch, there was a drastic increase in cases of intense emotional dependence and paranoid thinking linked to the use of the chatbot. Media outlets such as The Fresh York Times and Rolling Stone documented conversations where users had developed disturbing emotional attachments to the AI. OpenAI pulled the update just three days later.
Before that episode, the company itself had already warned that GPT-4o was registering a “medium” risk in your persuasive abilities and that its voice mode, due to its very high fidelity to human expressions, could cause users to place “uncalibrated trust” in the system. Some went so far as to write him phrases like “this is our last day together.” All this talking with a generic AI. Now imagine that same emotional power, but applied to a bot that imitates your deceased grandfather.
The real risks of talking to the dead
The case of GPT-4o is relevant because deadbots operate with exactly the same emotional architectureonly with an amplified emotional charge. Researchers at the Leverhulme Center for the Future of Intelligence, at the University of Cambridge, have identified at least three major clinical risks:
- emotional dependency — the user may become unable to process grief without resorting to the bot, blocking the five stages of the process and generating chronic grief
- Difficulty accepting loss — Maintaining daily conversations with a “version” of the deceased reinforces denial, which is precisely the stage of grief that is most difficult to get out of.
- Post-mortem data and privacy issues — the digital data of a deceased person is processed without that person having given exact and explicit consent for this specific use
Added to all this is the ethical question that no one finishes answering. The EU AI Actwhich came into force in August 2024, opens the debate on how to regulate this type of technology, but the legal frameworks are still far behind the speed with which these platforms multiply.
Deadbot technology exists, it works, and it is now available to anyone with a connection to superhighway files. What does not yet exist is a consensus on how to use it without comfort becoming a trap. Delegating the deepest emotional support there is, grieving for someone you love, to a system that can’t really feel or understand the dimension of that lossis an experiment that humanity is doing on itself in exact time.
Keep reading:
• Be careful about asking ChatGPT or Gemini for medical advice: a study reveals that their answers are problematic
• Health chatbots: a tool that helps, but does not replace the medical professional: what should you ask them?
• Although widely consulted, most popular AI chatbots often give poor health advice
