Police in the Indian state of Uttar Pradesh have arrested a man they say faked his own death with an AI-generated photograph in order to escape trial for murder. He had very nearly succeeded.
Rajkumar Rajbhar, 38, was arrested on Sunday at around 9am by Kotwali Nagar police, with help from the Special Operations Group and a surveillance team, near his home area of Khwaja Jahanpur. He had been out on bail in a case concerning a man shot dead.
What police say he did
Rajbhar was on trial over the 2016 killing of Yogesh Singh, who was shot near the Munsipura railway line on 19 December 2016. The case had been brought against Rajbhar and his friend Sudarshan Rajbhar on a complaint by the dead man’s nephew, Avinash Singh.
According to police, on 30 May 2025, sitting in Ghazipur, Rajbhar planned the staging of his own death. An AI-generated image was used of a body lying on a funeral pyre, with a superimposed face. Beneath the shroud, police say, was the carcass of a goat, which was cremated in place of a human body. Rajbhar also assembled photographs of the funeral and the procession to the cremation ground.
On the strength of those documents, a death certificate was issued by the Mau Municipal Council. He submitted it to the court, claiming he would no longer be alive to face trial, and then lived openly for about eighteen months.
The motive police describe
Additional Superintendent of Police Anup Kumar said Rajbhar, who failed class 12, had worked as a stock market trader and had collected money from prominent people in several cities on the promise of investing it in shares.
Police say the plot was intended to do two jobs at once: evade the legal process in the murder case, and avoid returning the money he had taken.
Because Rajbhar is no longer “dead”, the original murder trial will resume. A separate case has been filed against him for forgery.
The generated image is not the weak point
It is tempting to read this as a story about how good image generators have become. It is not, quite. Producing a plausible photograph of a shrouded body on a pyre has not been difficult for some time, and a face superimposed on a corpse is a forgiving task for a model — the subject is lying still, partly covered, photographed in poor light, in a setting where nobody expects a sharp portrait.
The weak point is the process. A death certificate is a foundational document: it closes bank accounts, transfers property, ends court proceedings, releases insurance. The chain here ran photograph to certificate to court filing, and the only verification at the first link appears to have been that the photographs looked right.
That chain was fragile before generative tools existed; people have been producing false death certificates with staged photographs and cooperative witnesses for a very long time. What has changed is the cost. Staging a cremation previously required a body or a convincing substitute plus people willing to lie about it. Now one of those two requirements is a prompt.
What would actually have caught it
Not image forensics. Detection tools are improving — Google opened its AI-detection portal to everyone this month — but they only reliably identify images carrying a watermark from a participating generator, and a municipal clerk is not going to run a forensic check on a funeral photograph.
What catches this is boring administrative design. A death certificate issued against a medical certification of cause of death, or a cremation-ground register entry, or any record created by someone with no stake in the outcome, breaks the chain immediately. The fraud worked because every input came from the person who benefited.
That is the transferable lesson for anyone building or auditing a records process here, from land registries to insurance claims to university certificates. The question to ask of any document is not whether the photograph attached to it is real. It is whether any part of the file was produced by somebody other than the applicant.
Related: an AI test agent that sent police a fabricated murder tip — the same problem of generated material entering a system built to trust what it is handed.
Prothom Alo credits NDTV for the report.
