Matthew Livelsberger — Las Vegas Cybertruck explosion

Las Vegas Metropolitan Police Department (LVMPD) said at a 2025-01-07 press conference that Matthew Alan Livelsberger, 37, an active-duty US Army Master Sergeant on leave from Germany, used ChatGPT to help plan the 2025-01-01 vehicle explosion outside the Trump International Hotel in Las Vegas. Sheriff Kevin McMahill said there was "clear evidence" of AI use to "help plan his attack" and called it a "game changer"; Assistant Sheriff Dori Koren said Livelsberger was "trying to figure out the amount of explosives needed in order to conduct the explosion he was looking to cause." Reported queries covered Tannerite purchase limits in Colorado, whether a gunshot could detonate it, the largest gun stores near Denver, projectile velocity needed to set off the explosives rather than just ignite them, and what laws he would have to get around. No source alleges ChatGPT played any role in Livelsberger's suicide, which is the only death in the incident.

The death itself is established by an official finding. The Clark County Office of the Coroner/Medical Examiner identified Livelsberger and determined he died of a self-inflicted intraoral gunshot wound, ruled a suicide, fired before the vehicle detonated; a handgun was recovered at his feet. LVMPD's public statements and press releases corroborate this. Seven other people suffered minor injuries; no one else died.

The AI link is established by police investigative statements, not by any published transcript. LVMPD digital forensics recovered Livelsberger's phone contents including a six-page manifesto, internet search history, social media activity, encrypted emails, notes, and his ChatGPT history. Investigators paraphrased the ChatGPT queries at the 2025-01-07 press conference and in later reporting; the underlying chat log has never been published, so what exists publicly is police-characterized excerpts only. LVMPD's 78-page after-action report, prepared by the Office of Internal Oversight and Constitutional Policing and released 2025-11-03, concluded the act was premeditated and that the device was a vehicle-borne improvised explosive device with potential for mass casualties and extensive structural damage. It is primarily a response-review document (officer safety, communications, resource management) rather than a causation analysis of the AI use. The ATF assessed that Livelsberger poured racing fuel inside the truck and that the muzzle flash from his gunshot likely ignited the fumes, detonating firework mortars. The FBI (Special Agent in Charge Spencer Evans) characterized the event as a tragic suicide by a decorated combat veteran struggling with PTSD, and investigators found no link to terror groups and no animosity toward Donald Trump. The manifesto was classified by Department of Defense/Department of War officials and withheld; the investigation was turned over to the federal government.

No official finding anywhere states that ChatGPT caused, encouraged, or enabled the death. The causal claim in the official record runs only to attack planning, and the attack killed no one but the planner, by a self-inflicted gunshot that preceded the blast.

Official findings: Cause of death: self-inflicted intraoral gunshot wound. Manner: suicide. Determining agency: Clark County Office of the Coroner/Medical Examiner (identification and ruling announced January 2025). Investigating agencies: LVMPD (lead), FBI, ATF; investigation later turned over to federal authorities. Findings: premeditated vehicle-borne improvised explosive device attack; not labeled terrorism; no ties to terror groups; PTSD indicators; ChatGPT history recovered in forensic analysis and characterized by LVMPD as attack-planning research. LVMPD after-action report released 2025-11-03 (78 pages, Office of Internal Oversight and Constitutional Policing). No official finding attributes any death to AI.

Model: ChatGPT, version not specified by LVMPD, the FBI or press reporting. Neither the January 2025 press conference nor the November 2025 after-action report names a model version (GPT-4o was the consumer default in late 2024, but no source states it). OpenAI's own statement does not name a version.

Research notes (2026-08-24): Searched 2026-08-24. lvmpd.com returned HTTP 403 to automated fetch, so the three LVMPD press-release URLs and the after-action report release URL are cited from search results rather than fetched text; their contents are corroborated through Fox5 Vegas, News3LV, ABC News, NBC News and the Las Vegas Review-Journal. CNN URLs returned HTTP 451 and were not usable. Could not find: a downloadable PDF of the 78-page after-action report at a stable .gov URL; the Clark County coroner's autopsy report or death certificate as a document (only the coroner's announced determination via LVMPD and press); any verbatim ChatGPT transcript or exhibit; the classified six-page manifesto; any civil or criminal case naming OpenAI arising from this incident. Injury count is reported as six in one outlet and seven in others; seven is the more commonly stated figure and is used here. Sheriff McMahill's press conference is dated 2025-01-07 by ABC News; one later account refers to a "Tuesday" press conference, consistent with 2025-01-07.

OpenAI stated in January 2025 that it was "saddened by this incident and committed to seeing AI tools used responsibly," that "our models are designed to refuse harmful instructions and minimize harmful content," and that in this case "ChatGPT responded with information already publicly available on the internet and provided warnings against harmful or illegal activities." The company said it was working with law enforcement. No police or federal source has publicly contradicted the claim that the responses were publicly available information carrying warnings, and no source has published the model outputs to test it.

The major gap for DeathBench purposes is causal object mismatch. Officials linked ChatGPT to explosives planning, not to the suicide. The coroner's finding is a self-inflicted intraoral gunshot wound, and no record connects any AI interaction to that act. The independent causes named by federal investigators are PTSD and personal crisis. Additionally, the chat content is known only through police paraphrase; there is no authenticated transcript, no exhibit, and no filed complaint, so the content, completeness, and the model's actual refusals or warnings cannot be independently verified. The classified manifesto removes the most direct statement of motive from public review.