Emerging Patterns in AI-Assisted Domestic IED Construction
Organization
Black Eagle Group™
Operational Domain
Threat Intelligence
Subject Vector
Adversarial AI / IED
Executive Summary
Classification Note
The Las Vegas incident is included for capability assessment as the first domestic case of AI-assisted IED construction. Per FBI records, it is classified as a suicide/stunt, not terrorism. Platform identities are generally withheld to maintain neutral reporting, except where specific applications (e.g., ChatGPT in INC-001) are explicitly named in official law enforcement forensic reports or public press releases.
This document provides initial documentation of three confirmed U.S. incidents in which law enforcement records — including FBI criminal complaints and DOJ indictments — confirm that generative AI applications were consulted during the research, design, or construction of improvised explosive devices (IEDs).
All three incidents occurred within a span of just over five months between January 1 and June 5, 2025. Caution is warranted: three cases over five months represent an insufficient sample to support broad conclusions about threat trajectories. This document should be read as an emerging pattern requiring systematic investigation rather than an established threat assessment.
Las Vegas Metropolitan Police Department Sheriff Kevin McMahill publicly described the first confirmed domestic case in January 2025 as a "game changer," characterising it as the first known incident on U.S. soil in which an AI application was used to assist an individual in device planning.
Primary Assessment
- Three confirmed incidents in 5 months
- Diverse actor profiles (Veteran to Civilian)
- AI as a planning and procurement tool
- Mental health factors in 2/3 cases
- Human agency remains primary driver
Analytical Limitation
Additional data collection is required before reliable threat evaluation or policy recommendations can be supported.
Section I — Confirmed Incidents
Las Vegas VBIED — AI-Assisted Explosive Planning
Impact
MEDIUM
Date
January 1, 2025
Classification
AI-ASSISTED IED CONSTRUCTION (Suicide / Stunt — FBI)
Actor Profile
Matthew Alan Livelsberger, 37 — U.S. Army Special Forces Master Sergeant (active duty). Decorated combat veteran with five Bronze Stars. Nineteen years of service. FBI assessed he was likely suffering from PTSD.
AI Tool
ChatGPT — Confirmed by LVMPD
Device
VBIED — Tesla Cybertruck with 60lbs pyrotechnics
AI Use Case
Queries included: quantity of explosives required for desired effect; legality and purchase locations for fireworks in Arizona; bullet velocity thresholds for detonation versus ignition.
Palm Springs ANFO Car Bomb — AI-Assisted Explosive Chemistry
Impact
HIGH
Date
May 17, 2025
Classification
DOMESTIC TERRORISM (FBI / DOJ formal designation)
Actor Profile
Guy Edward Bartkus, 25 (deceased) & Daniel Jongyon Park, 32. Bartkus left a manifesto articulating anti-natalist / pro-mortalist ideology. FBI characterized ideology as nihilistic.
AI Tool
Unnamed AI chat application
Device
ANFO car bomb: ~270 lbs ammonium nitrate
AI Use Case
Researched how to produce powerful explosions using ammonium nitrate and fuel, with queries focused on explosive composition, diesel and gasoline mixture ratios, and detonation velocity.
Manhattan IED Deployment — AI-Assisted Device Construction
Impact
HIGH
Date
ARREST: June 5, 2025
Classification
ATTEMPTED DESTRUCTION, IED POSSESSION, TRANSPORT
Actor Profile
Michael Gann, 55. No military or formal explosives background. Prior criminal record: 30 total arrests; 28 total convictions. Acted entirely alone.
AI Tool
Unnamed AI application (self-reported)
Device
Flash powder IEDs: potassium perchlorate & aluminium
AI Use Case
Gann self-reported that he used an AI application to determine which chemicals to purchase and how to mix them. The USAO-SDNY indictment separately documents conventional internet search queries for flash powder formulation.
Section II — Incident Patterns
Operational Assessment
Execution Model: AI functioned strictly as an auxiliary research tool across all incidents. All actors independently executed the physical procurement and construction phases.
Actor Typology: There is no unified ideological driver (motives spanned personal grievance, anti-natalism, and anti-immigration). However, documented mental health instability was a factor in two of the three primary actors.
AI Functionality Divergence
Trained Actors (Livelsberger)
Leveraged AI for operational logistics, planning efficiency, and procurement parameters (e.g., legal thresholds, bullet velocity / ignition threshold queries).
Untrained Actors (Bartkus & Gann)
Leveraged AI for fundamental technical knowledge (e.g., chemical compositions, mixture ratios).
Section III — Limitations & Gaps
| Collection Gap | Description | Required Action |
|---|---|---|
| Baseline Rate Comparison | To measure true AI impact, collection must establish a pre-2022 baseline of IED construction rates by untrained actors using conventional internet searches. | Collate ATF Bomb Data Center public reports & federal explosives incident data (2018–2022) |
| The Interdiction Denominator | Current intelligence only reflects completed plots. There is zero visibility into individuals who queried AI for explosives but were blocked by safety guardrails. | Systematic PACER/DOJ monitoring for AI mentions in disrupted plots |
| AI Output Verification | Public records document the actor queries but omit the AI responses. It is unknown whether the platforms provided accurate blueprints or hallucinated data. | Retrieve court exhibits detailing exact AI session logs (Gann & Park prosecutions) |
Dataset Limitations
Statistical Insignificance
Three isolated cases over five months cannot support predictive threat modeling.
Attribution Dilution
The specific capability uplift attributable to AI cannot be isolated due to concurrent confounding factors (prior training, independent research, etc).