Should a social media account die with its owner? Meta has come to contemplate the opposite. At the end of 2025, the company was granted a patent that describes how a language model could be trained with a user’s historical data to continue posting, reacting and even responding to direct messages on their behalf when they were absent, or dead. Although the company states that it will not move forward with the idea, the document opens a debate on digital identity, grief and the business model of the platforms.
The patent registered in 2023 and granted later describes a system based on language models (LLM) capable of “simulating” the activity of a user on a social network. According to the text, the model could be used when a person takes a long break or even if the user has died.
The system would be trained with profile-specific data: posts, comments, likes, interaction patterns, and even private messages. The result would be a digital clone capable of interacting with third parties, reacting to content, commenting and, potentially, simulating audio or video calls.
From Meta they remember that they register patents to protect concepts, not necessarily to implement them
However, in statements after Business Insidera company spokesperson assured: “We have no plans to move forward with this example”. Thus, from Meta they remember that they register patents to protect concepts, not necessarily to implement them. Although the simple fact that the idea was formulated in technical terms reveals the extent to which big technology companies have explored new territories for generative models.
The border of the posthumous avatar
Meta is no stranger to the debate about digital legacy. Almost a decade ago, Facebook introduced the “legacy contact” figure, which allows you to designate a person to manage the account after death. But the step suggested by this patent goes further, keeping the profile active.
In an interview in 2023 with podcaster Lex Fridman, Mark Zuckerberg reflected on the possibility of virtual avatars allowing him to interact with deceased people. “If someone has lost a loved one and is grieving, there may be ways in which interacting or reliving certain memories can be helpful.”he stated then. Although he also recognized: “There is probably a point where it can become unhealthy.”
The patent document even notes that the permanent absence of a deceased user has a “more severe and permanent” impact on the experience of other users. That is, the platform perceives death as an interruption of the social flow.
And on platforms whose profitability depends on the time spent and the volume of activity, an account that stops publishing represents a potential decrease in engagement. And if the network progressively becomes a “digital cemetery” of inactive profiles, the erosion could be noticeable.
In the current scenario, Facebook accumulates millions of accounts of deceased users. At the same time, their feeds are filled with AI-generated content. Therefore, the border between human presence and automation is already blurred. In this scenario, a digital clone trained on historical data would prolong the symbolic life of a profile, keeping its contribution to the advertising ecosystem active.
Grief tech, between innovation and controversy
The category known as grief tech It has already generated start-ups that allow the creation of chatbots trained with data from deceased people. Microsoft patented a similar system in 2021, which it later discarded as disturbing. But companies like Replika or “deadbot” initiatives have tried to occupy that space, between innovation and controversy.
Digital estate planning begins to include specific clauses on AI
In parallel, some famous people have begun to legally shield their image and voice to prevent unauthorized uses after their death. So digital estate planning is beginning to include specific clauses on artificial intelligence.
Regardless of whether Meta develops this technology or not, the existence of the patent shows how far experimentation with generative models is going. The question shifts, therefore, from whether someone can technically be simulated, to whether it should be done. Because a model trained with a person’s digital history can reproduce their style, tone, and interaction patterns. But identity is not only a sum of data: it implies consciousness, context and evolution. Converting it into a system that responds automatically after death shifts the debate from the technological to the philosophical and legal fields and forces us to redefine what it means to die in the digital age.
