Wargaming tools for national security: who does what
National security wargaming runs on a few distinct models: facilitated analytic games built around a sponsor’s question, consultancy-run exercises, research experiments that pit people against language models, commercial simulations, and now scored single-player formats. They answer different questions, so the fit depends on what you need to learn, from whom, and how often.
How do the models compare?
| Model | Who runs it | Format | Who can take part | What it measures | Typical use |
|---|---|---|---|---|---|
| Government and FFRDC analytic wargaming | Federally funded research centers and labs for a government sponsor: Johns Hopkins APL’s GenWar Lab (and, as reported by Breaking Defense, a Strategic AI Gaming Engine), CNA, RAND. | Facilitated, multi-team, built around the sponsor’s question; months to prepare, days to play, an analysis after. | The sponsor’s invited participants. | Insight into the problem: options, assumptions and risks. | Concept exploration, force and posture choices, senior-leader tabletop exercises. |
| University simulation platforms | Academic centers. The University of Maryland’s ICONS Project builds simulations for education, training and research, including distributed multi-side and human–AI games. | Role-play simulations on a platform, run by an instructor or facilitator. | Classrooms, agencies, training consultants, researchers. | Learning outcomes; research data on how people decide. | Teaching, team training, decision-making research. |
| Consultancy-run exercises | Consulting firms and contractors, under contract to a government or corporate client. | A facilitated exercise designed for the engagement; the client sets the question and keeps the findings. | The client’s own staff. | Whatever the client scoped: readiness, a plan’s weak points, a decision rehearsed. | Strategy offsites, crisis rehearsal, resilience and incident exercises. |
| AI-versus-human experiments | Research institutes and hackathons. George Mason University’s National Security Institute ran a Taiwan Strait crisis with humans and GPT-4 in the same US role; the SCSP 2026 hackathon produced Kriegsspiel, an open-source engine with a language-model red cell. | A research protocol: one scenario played by people and by a model, every move logged. | Recruited participants and experts. | Where human and model behavior diverge. | Research on language models in crisis decision-making. |
| Commercial and hobby wargames | Publishers. Command: Modern Operations (Matrix Games) is an operational simulator with a professional edition on the same engine; Next War: Taiwan (GMT Games) is a two-player board wargame of a near-future conflict. | Rules, units and a map; play against the system or another player. | Anyone who buys them. | The outcome under the rules. | Hobby play, self-study of operations and campaigns. |
| WarPaths | WarPaths authors its own scenarios; think tanks, research organizations and consultancies author games from their own reports. | Single-player, turn by turn: read the situation, consult AI advisors, write your orders in your own words, and the world responds. Current scenarios run about six turns, at your own pace; turn count varies by scenario. | Individuals, by request during the beta; an organization’s invited audience. | The reasoning behind each decision, scored on five dimensions into one number. | Repetitions between the big exercises; a scored record of judgment; a report turned into a game. |
Where does WarPaths fit?
WarPaths does not replace a facilitated analytic wargame. A room of experts working a sponsor’s question for two days produces something different from what a single player at a screen can.
WarPaths makes three narrower claims. Repetitions: a facilitated game is an event most people get once or twice a year; a WarPaths game is a sitting you can take as often as the scenarios allow. A record: each game is scored on the reasoning behind its decisions, so individual judgment accumulates into something you can look back on. Reach: a report that would be read once becomes a game an audience plays, and the organization sees how its readers reasoned.
Which model fits which need?
- A sponsor’s question, with the people who own the answer in the room: a facilitated analytic wargame.
- A class or a team practicing a role together: a simulation platform.
- A hypothesis about how models decide: an experiment with a logged protocol.
- Operations and campaigns for their own sake: a commercial wargame.
- Practice between exercises, a score you can compare, or a report your audience should reason through: WarPaths.