Digital life after death – Meta has patented AI that posts on behalf of the deceased

Meta has been granted a patent for artificial intelligence that simulates the activity of deceased Facebook and Instagram users. Likes, comments, message replies – as if the person were still alive. Sounds like a Black Mirror episode? It’s a reality described in patent no. US 12,513,102.

What does Meta’s patent actually describe?

In late December 2025, Mark Zuckerberg’s company received a patent that has been buzzing through the tech industry – and beyond – for the past few weeks. The documentation describes a large language model (LLM) designed to simulate user behavior on Meta’s social platforms – during their extended absence or after their death (source: Business Insider).

The lead patent author is Andrew Bosworth, Meta’s CTO. The documents were filed in 2023, but the patent itself wasn’t granted until late 2025.

Patent details

• Patent number: US 12,513,102 B2
• Title: “Simulation of a user of a social networking system using a language model”
• Inventor: Andrew Garrod Bosworth (Meta CTO)
• Patent owner: Meta Platforms Technologies LLC, California
• Number of patent claims: 18
• Classification: H04L (digital communication), G06F (natural language processing), G06N (machine learning)

Patent timeline:

1. November 29, 2023 – patent application filed (no. US18/523,545)
2. January 23, 2024 – formal transfer of rights from Bosworth to Meta Platforms Technologies LLC
3. September 7, 2024 – international patent application filed (PCT/US2024/045744)
4. May 29, 2025 – first publication of the application (US20250175448A1)
5. August 22, 2025 – decision to grant the patent (Notice of Allowance)
6. December 30, 2025 – patent officially granted (US12513102B2)
7. November 29, 2043 – projected patent expiration date

What is such a model trained on?

The system is built on a specific user’s data. The patent lists several types:

• posts and comments – writing style, favorite topics, the way they react to content
• likes – preferences, interests, activity patterns
• photos and video – visual context
• voice messages – tone, speaking patterns, characteristic phrases

From this, the AI builds something like a “digital personality” capable of predicting how a given user would react to specific content.

What can such a bot do?

The patent doesn’t stop at liking. The capabilities described include:

• reacting to friends’ content (likes, comments)
• replying to private messages
• generating new posts in the user’s style
• in the future – simulating audio and video conversations

Meta’s position

A company spokesperson assured Business Insider that Meta has no plans to roll out this technology and that the patent is conceptual. Patents are often filed to protect ideas that may never become products (source: Businessinsider.com).

The thing is – the very fact of patenting such a detailed concept, with described architecture, model training methods, and use case examples, shows that someone at Meta put far more than a few minutes of brainstorming into this project.

A “whistleblower”* warned back in 2023

• * Whistleblower – someone who exposes information about activities considered improper or illegal within an organization

The patent didn’t appear out of thin air. In 2023, an anonymous post appeared on 4chan from someone claiming to work on this project at Meta (codenamed “Project Lazarus”).

The author of the leak described a system meant to create plausible posts as if the user were still alive. They mentioned features like aging up profile photos. And they shared their concerns:

At first we were told this would be a service for people grieving the loss of loved ones. It seemed like a decent idea. But things took a strange turn.

The whistleblower also pointed out something unsettling: how little input data is needed for AI to start convincingly mimicking someone’s communication style.

At the time, no one could verify any of it. The December 2025 patent confirmed that the leak was telling the truth.

Why does digital resurrection raise concern?

Zuckerberg himself admitted in a conversation with podcaster Lex Fridman that interacting with a digital version of a deceased loved one could help with grief – but it could just as easily turn out to be “unhealthy”.

You can listen to that specific podcast here:

(You’ll hear this point around 33:30)

“If someone has lost a loved one and is grieving, there may be ways in which being able to interact or relive certain memories could be helpful. But then there’s also probably an extent to which it could become unhealthy.”

“We have, you know, a fair amount of experience with how to handle death and identity and people’s digital content through social media already, unfortunately.”

Researchers at the University of Cambridge went further. They published an analysis in Philosophy and Technology on the potential risks tied to so-called deadbots – chatbots imitating the deceased.

Digital “haunting”

Among the risks, they listed a scenario where the bot sends notifications and messages to loved ones, creating the impression that the deceased is still around. Even if someone initially finds comfort in such interaction, over time it can become an overwhelming emotional weight.

The researchers also warned of a scenario where a deadbot starts serving loved ones ads in the deceased person’s communication style – a kind of digital stalking from beyond the grave (source: University of Cambridge).

The consent problem

Dr. Tomasz Hollanek, co-author of the Cambridge study, stressed that digital afterlife services should account for the rights of both sides: those being “recreated” and those who will interact with the simulations (source: Cybernews).

In 2025, deepfakes appeared showing Martin Luther King Jr. speaking racist words. Zelda Williams, daughter of the late Robin Williams, publicly asked fans to stop sending her AI-generated videos featuring her father (source: Cybernews).

Now imagine a situation where it’s not fans, but a platform’s algorithm generating such content automatically. Without asking anyone for permission.

The business motive behind digital immortality

Meta’s patent officially talks about caring for the experience of users who feel a friend’s absence. But there’s a second layer.

A simulated user is still a user. They generate content, collect interactions, feed data to the algorithms. More automated activity means more material to train future AI systems on – and more time spent on the platform by the living users.

Compare the two approaches Meta takes toward post-mortem accounts:

• “Legacy contact” (current feature) – the profile becomes a memorial page, no one pretends the user is alive. Conservative, respects the fact of death.
• The AI bot patent (concept) – active simulation of presence. The bot reacts, comments, posts. The user “lives” on the platform.

That’s a completely different scale. The patent was registered in the US patent system under the “psychology” category – which raises questions about intent on its own.

As one commentator put it: at some point, being alive will simply become another data point in the algorithm (source: Vice).

Who tried this before?

Meta isn’t the first company to reach for the idea of digital resurrection. The history of attempts goes back nearly a decade – and each one carries different lessons.

The Roman chatbot – the first high-profile attempt (2016)

In November 2015 in Moscow, Roman Mazurenko – a 30-year-old entrepreneur and a known figure in Moscow’s art circles – was hit and killed by a car. His close friend, Eugenia Kuyda, spent weeks after his death reading thousands of texts they had exchanged since 2008. Mazurenko wrote in a distinctive way – he struggled with dyslexia and used unusual phrasing. Those messages were one of the few tangible traces left of him.

Kuyda ran an AI startup called Luka. She had the idea of training a neural network on those messages so it would respond in Roman’s voice. She gathered messages from ten close people, including Mazurenko’s parents. That’s how the chatbot @Roman was born – the first widely covered case of “digital resurrection” (source: The Verge / The Week).

Reactions from his close ones? Mixed.

• Roman’s mother said the bot helped her get to know her son better – she learned what he thought about various things.
• His father took it differently: hearing a computer program use his son’s phrases was painful, and the wrong answers amplified the awareness of the loss.
• Four friends refused to interact with the bot at all.
• Many others found the resemblance unsettlingly accurate. One described it like this: the phrases the bot uses are really his phrases. You can feel that he would have said it that way.

(sources: IBTimes, MIT Press Reader)

Kuyda later expanded Luka into the app Replika – a chatbot that learns to mimic the user’s personality. Replika still has millions of users worldwide.

Microsoft’s patent (2021)

In 2021, Microsoft was granted a patent for a chatbot that, based on social media posts, messages, and voice recordings, could simulate a conversation with a specific person – alive or deceased. The patent even mentioned celebrities and historical figures.

Tim O’Brien, then Microsoft’s general manager of AI programs, publicly summed up the matter in one word: “disturbing”. He confirmed the company had no plans to build a product on the patent. He added that the 2017 application predated the ethics review process Microsoft has been running since (source: CNN).

Patent US 10,853,717 B2

• Title: “Creating a Conversational Chat Bot of a Specific Person”
• Inventors: Dustin I. Abramson, Joseph Johnson Jr.
• Owner: Microsoft Technology Licensing, LLC
• Filed: April 2017
• Granted: December 1, 2020 (media wrote about it in January 2021)
• USPTO: uspto.report/patent/grant/10,853,717

HereAfter AI – memories instead of simulation

A different path was taken by James Vlahos, an American journalist. When his father was diagnosed with lung cancer, Vlahos started recording their conversations. After 12 sessions, an hour and a half each, he had nearly 92,000 words transcribed across 203 pages. Anecdotes, songs, childhood memories. From this material he built a chatbot he named “Dadbot”.

When he published an article about it in Wired in 2017, requests started pouring in from around the world. People sick or caring for dying loved ones asked: will you make me a Mombot? Will you make me a Dadbot? (source: The Daily Beast)

Vlahos turned the idea into a startup, HereAfter AI. The platform lets users record life memories by answering guiding questions. After death, loved ones can “talk” to an avatar that responds based on the previously recorded stories.

The difference from Meta’s patent is fundamental. HereAfter AI relies on consciously recorded memories and the consent of the person sharing them. Meta’s patent describes automatic cloning of social media activity – without the user’s involvement.

Other industry attempts

• Amazon Alexa (2022) – a demo of mimicking a deceased grandmother’s voice from less than a minute of audio recording. Never launched.
• Eternos – digital versions of users
• StoryFile – interactive video interviews with AI
• DeepBrain AI (Korea) – video avatars with 96.5% similarity to the original
• Project December – text chatbots simulating the deceased under the slogan “Simulate the dead”

Where does the line run?

The law isn’t keeping up with the technology. There’s no clear regulation around digital identity after death. Who owns the “digital self” – the heirs, the platform, no one? Can someone create a digital clone of you without your consent?

The Cambridge researchers call for safety standards in the digital afterlife industry. They point to three pillars:

1. Transparency – the user must know they’re talking to a bot, not a person
2. Consent – both from the person being “recreated” and from those who will interact with the simulation
3. Right to deletion – the option for loved ones to switch off the deadbot at any time

Meta’s patent, regardless of whether it ever becomes a product, sets a direction. The technology to create credible digital copies of people exists. The question is no longer “is this possible”, but “who decides whether it should exist – and on whose terms”.

What about you? Would you want your Facebook profile to “live on” after your death? Or would you rather make sure no one creates a digital clone of you without your consent?

FAQ