Contested Digital Legacy: Who Owns Our Artificial Ghost?

What happens to our data when we die? Today, the chats and emails of a deceased person can be used to train a "deadbot," an Artificial Intelligence capable of i

The persistence of digital identity. Source: StockCakeDeath, in the digital era, has ceased to be a clear boundary. When an individual passes away, today they do not leave only real estate, bank accounts, or physical assets to be divided. They leave an enormous, disorderly trail of data, memories, works and, increasingly often, a simulated presence capable of continuing to produce new messages. The so-called deadbots (or griefbots), Artificial Intelligence models trained on chats, emails, voice notes, and videos of a deceased person to simulate their communication style, are forcing legislators to confront an unprecedented question: who owns the "digital ghost"?

Seeking a simple answer to this question is an illusion. The deceased does not leave a single "digital asset" to be assigned to heirs as a whole. The creation of a deadbot splits identity into fragments subject to profoundly different jurisdictional rules: the data used for training, access to accounts, the replication of image rights, up to liability for the new textual content generated by the machine.

In this in-depth analysis for the Scenarios and Reflections column, we will dissect the legal and ethical infrastructure of digital inheritance. We will discover why establishing "who owns the deadbot" is only half the question. The other half, far more complex, concerns who owns the right to authorize its creation and what protections remain for living people whose secrets have remained entangled in the training material.

1. The European Regulatory Void and the Italian Exception

To untangle the knot, one must start from the European legal paradox. Many believe that the General Data Protection Regulation (GDPR) protects individuals beyond death. The reality is the opposite: Recital 27 of the GDPR explicitly establishes that the regulation does not apply to the personal data of deceased persons [1748]. The European Union has chosen not to legislate on the matter, leaving Member States the option to adopt national rules. This generates a void and profound jurisdictional fragmentation.

In this scenario, Italy represents a protective exception. The Data Protection Authority, citing Article 2-terdecies of the Privacy Code, has established that certain persons may exercise specific rights over the deceased's data [1769, 1770]. This power is not universal, but is granted to those acting in their own interest, to protect the data subject, as an agent, or for "family reasons worthy of protection."

However, Italian law introduces a fundamental counterbalance: the deceased may have previously prohibited in writing the exercise of these rights on a specific digital service. Heirs do not automatically become "owners" of the deceased's personality; they become, under specific conditions, the custodians of the data.

2. Anatomy of a Deadbot: Separating Data from Identity

When a company or a family member decides to create a post-mortem avatar, they clash with a fragmentation of rights. The ELI Model Rules on Succession and Access to Digital Remains, proposed by the European Law Institute in 2026, offer a crucial distinction to frame the problem [1804, 1805]:

  • Digital assets with economic value (cryptocurrencies, monetized accounts) can and should fall under ordinary patrimonial succession.
  • Personal "digital remains" (chats, emails, voice messages) are indissolubly linked to identity, privacy, and dignity. They should not be treated as mere inheritable property, but governed by rigidly protected access.

A further level of complexity is given by copyright and intellectual property. A report by the U.S. Copyright Office ("Copyright and Artificial Intelligence, Part 1: Digital Replicas") separately analyzes the digital replication of voice or visual appearance with respect to copyright on texts written by the person during their lifetime [1782, 1798]. Training a deadbot means using the works written by the deceased (often covered by copyright), synthesizing their voice (image and personality rights, variable from nation to nation), and generating new texts that have no recognizable human author.

3. The Rights of the Living and the Ethics of Digital Ghosts

If the legal framework is fragmented, the ethical one outlined by academic research is fortunately converging toward rigorous standards. The essay published in Philosophy & Technology ("Griefbots, Deadbots, Postmortem Avatars…") and the study in Ethics and Information Technology ("The Making of Digital Ghosts") delineate the boundaries of responsibility for the digital afterlife industry [1746, 1756, 1758].

The real problem of a deadbot is not only the consent of the deceased, but the confidentiality of third parties. The material needed to train an AI (years of WhatsApp or Gmail archives) inevitably contains communications from living people: confidences, secrets, health or financial data. When an heir uploads this archive to a technology company's servers to "resurrect" the relative, they are exposing the data of dozens of unaware and living third parties, who retain all their rights intact under the GDPR.

Scholars propose an ethical framework based on non-negotiable pillars:

  1. Intent expressed during life: The creation of a bot should require explicit prior consent from the deceased, not a simple "non-opposition."
  2. Absolute transparency: Anyone interacting with the system must know at every moment that they are talking to a neural network, not a spirit.
  3. Agency limits: The system must not be able to enter into contracts, express new political opinions, or attribute to the deceased intentions they never manifested during life. Generative AI is a stochastic machine that "hallucinates" responses; making it hallucinate a dead person's opinion is a direct violation of their posthumous dignity.

Key Operational Takeaways

  • Plan Your Digital Will: Do not rely on Big Tech's standard policies. Use "inactive account manager" settings (available on platforms like Google or Apple) and draft a written document in which you explicitly state whether you prohibit heirs from using your data for training generative systems.
  • Segment Ownership: Estate lawyers must begin to distinguish access to accounts for the recovery of financial assets from the massive export of textual data for third-party uses, specifying separate prohibitions.
  • Limit Access to Bots: Platforms offering digital afterlife services must implement sandboxing systems, ensuring that the bot can only be queried by a restricted circle of authorized family members, preventing public spectacle or commercial use of the digital replica.

The deceased of the 21st century does not leave as an inheritance a closed and static digital urn. They leave a potential engine, a dataset capable of fueling an active and reactive simulation. Reducing the management of this presence to a patrimonial dispute between heirs and technology companies means misunderstanding the very nature of generative technology. The deadbot is not a recording of the past, but an artificial actor thrown into the present. Deciding who has the right to turn it on, and above all to turn it off, will be one of the greatest challenges for civil law and bioethics in the coming decades.

Bibliographic References and Sources

Article by the Editorial Team of La Bussola dell'IA