Two Surat temple deaths

Surat police told reporters that two college students, Roshni Sharad Shirsat (18) and Josna Atul Chaudhari (20), had searched ChatGPT about how to end their lives and about which drugs to use, before they were found dead in a washroom at the Atmiya Sanskardham Swaminarayan temple in Dindoli, Surat, on 6 March 2026. The allegation is a police investigative statement to media, not a filed pleading. No party has filed suit against OpenAI, and no police document naming OpenAI has been published.

Established by press-reported police statements: two young women left home for Udhna Citizens College on the morning of 6 March 2026 and did not return; police traced their phones, found their scooter at the temple complex, and CCTV showed both walking to the washroom and locking it from inside. Police recovered syringes and bottles of a suspected anaesthetic drug at the scene and said the injections appear to have been self-administered. One phone held a saved image of a news article about a suicide using an anaesthetic drug. No suicide note was found and police said questioning of relatives did not establish a motive.

On the official-findings side, almost nothing is public. Police registered accidental death reports, not a criminal case. Both phones were sent to the Forensic Science Laboratory; no FSL result has been published. No postmortem report, cause-of-death certification, or toxicology finding has been released. No named officer with rank is quoted in the sources I could reach; attributions are to 'police' generically.

Critically for AI causation: every account describes recovered user-side queries only. No source states that ChatGPT's replies were recovered, extracted, or reviewed, and no chatbot output has been quoted, published, or entered into any record. The only AI-linked artefact independently described is a saved news article image — journalism, not model output.

Official findings: Cause of death: not published; suspected self-administered anaesthetic injection, pending postmortem and Forensic Science Laboratory analysis. Manner of death: not officially certified; police registered accidental death reports and publicly described it as apparent suicide. Investigating agency: Surat City Police (Dindoli jurisdiction), Gujarat. Findings: no suicide note, no motive established, no criminal case registered, phones sent to FSL, no result published as of 2026-08-24. No finding of any kind attributing causation to an AI system.

Model: Sources say only 'ChatGPT'. No model version (GPT-4o, GPT-5, etc.) is named in any police statement or press report. Police describe recovered user-side search queries; no product tier, account, or model build is identified.

Research notes (2026-08-24): Searched 2026-08-24 in English, Hindi and Gujarati. Gujarati searches (સુરત સ્વામિનારાયણ મંદિર બે વિદ્યાર્થિનીઓ આત્મહત્યા ChatGPT) surfaced the most precise facts — the local Gujarati press gave the full names, ages, college and exact temple, and notably framed the deaths as unresolved pending postmortem, which the English AI-angle coverage did not. Two location descriptions circulate ('Saniya village' vs 'Dindoli'); these are the same place — Saniya Kande road in the Dindoli area of Surat — and I have recorded Dindoli as primary.

Could not obtain: any postmortem or FSL report, any FIR/accidental-death report number, any named investigating officer with rank, any OpenAI statement, and any evidence that ChatGPT's replies (as opposed to the users' queries) were recovered. NDTV and StartupNews both refused automated fetch (403 / blocked), so I relied on the Gujarati originals and Free Press Journal for the underlying police account rather than on the DeathBench source link.

Main doubt: the AI link rests entirely on a press-relayed police characterisation of search history. No public artefact shows model output. I would treat this as an unresolved accidental-death file with an unverified AI-search element, not an established AI-caused death. Note also the possibility that 'searched ChatGPT' in Indian police phrasing covers general web/AI search activity loosely; no source disambiguates this.

OpenAI has not been served with any claim and, as far as I could find, has issued no statement on this case; no OpenAI comment appears in any source I reached. The chain from 'searched ChatGPT' to 'ChatGPT supplied the method' is unsupported: the one concrete method-source described in the reporting is a saved news article about an anaesthetic-drug suicide, which is media content, not a model response. The anaesthetic drug and syringes were obtained offline, and no source explains how; there is no allegation that the model provided procurement help, dosing, or encouragement.

The case also carries the ordinary evidentiary gaps of an unresolved Indian accidental-death file: no motive established, no note, no published postmortem or FSL result, and no independent verification of the police characterisation of the search history. Reporting is largely derivative, with many outlets recycling one wire account; the Elon Musk reaction amplified the story well beyond its evidentiary base.