When American comedian “The Amazing Johnathan,” also known as John Edward Szeles, died in 2022, his wife became an early client of a company offering artificial intelligence resurrections. Anastasia Synn, described as a “cyborg magician,” turned to You, Only Virtual.

The company aims to create “Versonas,” artificial intelligence personas built from messages and recordings produced while someone was alive. Founder Justin Harrison developed the idea after his mother was diagnosed with terminal cancer.

With her permission, Harrison assembled recordings and interactions that could recreate her likeness digitally after death. In the BBC documentary AI Confidential with Hannah Fry, he demonstrated technology designed to mimic speech patterns and enable a kind of phone conversation with someone who has died.

Digital recreations of the dead are not limited to private bereavement. The late Ian Holm controversially appeared in Alien: Romulus, while other technology has been used for performances featuring Whitney Houston singing Queen songs and Freddie Mercury singing Whitney Houston songs.

Artificial intelligence has also entered political advocacy and the courtroom. Joaquin Oliver, a student killed in the 2018 Parkland school shooting, was recreated for a project led by his father in which Oliver gave testimony to Congress supporting stricter gun control laws.

A 2025 study published in New Media & Society reviewed 50 “AI resurrection” projects and grouped them into three categories. Mundanization transforms ordinary people into synthetic personas, while spectacularization turns deceased cultural figures into performers.

The third category, sociopoliticization, involves recreating dead people for political advocacy, witnessing and commemoration. Although each category presents distinct ethical concerns, the projects share the possibility that someone’s identity can remain socially and economically productive after death.

Consent sits at the center of the problem. “One of the most striking things we encountered was how easily the dead can be made to participate in activities they never agreed to while they were alive,” study author Tom Divon told IFLScience.

Divon said decisions about recreating deceased people often rested with surviving relatives, companies, cultural institutions or political actors. The person being recreated could no longer approve those decisions or withdraw from them.

The study did not determine how much consent is legally required within the grief technology sector. It did, however, identify the absence of meaningful consent as a recurring ethical concern, particularly when personal conversations, photographs and voice notes are used to produce a digital persona.

“Personally, I think we should begin by asking what people explicitly authorised while alive,” said Divon, “rather than assuming their digital remains become freely available for others to use after death.”

The study’s authors proposed the term “spectral labor” for situations in which dead people are compelled to haunt the present for contemporary purposes. Those purposes may include comforting grieving people, filling a political role or performing songs they never sang while alive.

Professor Ines Testoni of the University of Padova, a death education expert, told IFLScience that these recreations can let people avoid one of grief’s greatest difficulties, separating themselves from those they have lost. Yet the technology also allows a person’s image, voice and persona to circulate without their continuing control.

Divon pointed to Charlie Kirk, an American right wing political activist who was shot dead in 2025 at age 31. Recreations of his voice and image generated by artificial intelligence circulated online, including fabricated messages that made him appear to continue addressing supporters.

Some of those recreations appeared as tributes, while others included explicit political and religious messages. Divon warned that such material can extend a person’s political authority after death and allow other people to manufacture statements in that person’s name.

Even uses presented as morally justified leave difficult questions about whose voice is actually being heard. “The deceased person's? The family's? The activist's? The technology company's?” asks Divon.

These projects use generative artificial intelligence and large language models to produce fresh responses from existing data and imitate familiar speech patterns. They cannot develop a new idea, change their minds or consent to claims about what a dead person would have wanted.

As the technology grows more capable, the central issue is no longer whether a convincing recreation can be built. It is who controls the resulting persona, whose interests it serves and whether the person being imitated ever wanted to return at all.