"Endgame" — a corporate and geopolitical AI wargame

The US Congress levies a tax on all non-DOD contracted AI Labs. In response, several US labs open branch offices in China out of protest. An enraged America quietly hires its most loyal tech firms to cyberattack Chinese power infrastructure, causing a crisis and locally spiking compute costs for Chinese labs. In this moment of political chaos, what choice does the CCP have but to bow down to the hardliners and kinetically challenge a US Navy Carrier Strike Group?

Endgame (working title) is a Megagame inspired by the AI safety roleplay exercises of D. Scott Phoenix et. al. and the British "Megagame" tradition, which goes back to the 1970s. Key stakeholders in the global AI ecosystem are starving for rigorous, well-designed rulesets to help important people understand the impacts of their actions for the future.

I am designing a game to solve such a problem.

In April 2026 I played D. Scott Phoenix's "The Endgame", a fascinating exercise in AI simulation in which a large number of players took on the roles of various interest groups and simulated the near future. It was a roleplay; a sort of game master (GM) mediated theatre improv exercise in which players took on the roles of AI labs, major world governments, and auxiliary bodies, as well as "The AI" in its growing power and sentience. Player teams would state their actions ("OpenAI release a new model!" or "China imposes new sanctions on the European Union!") and the GM would resolve them.

This idea is nothing short of brilliant. But it is flawed; it lacks rigor. Phoenix's game made no effort to model resources, capital, political power, warfare, the markets, natural resources, or anything else. Additionally, Phoenix's game requires an AI expert (such as Phoenix) to sit in the GM's chair and make in-game rulings based upon his expertise. Such a dependency does not scale well.

Phoenix's idea is excellent, but it needs something more. It needs a game designer. This is where I come in.

I propose to spend the next several months developing a Megagame based on Phoenix's ideas. I will develop and test a ruleset for 30 to 60 people to facilitate a game lasting 4 to 8 hours. My game will place player teams into the hotseats of the AI decision space— everything from the VP of Technology for a minor AI lab to the President of the United States— and it will include rigorous rules for everything from LLM model training to mergers & acquisitions to sea-to-air combat.

The hardest of this work will be done from my desk in Toronto, writing out design docs, value tables, and component designs. The hardest (and most capital intensive) parts of this project will be playtesting. Dr. Dillon Burke points out that the only way to test a Megagame is to play a Megagame, and testing a 30-60 player game has non-negotiable fixed costs.

If fully funded, I will design a rigorous but battle-hardened ruleset for an AI Safety megagame, simulating global economics and geopolitics for the coming years. I will use the funds not only to organise game sessions but to allow other institutions to use my rules to run the game anywhere. I will be able to deliver rules, game components, and guidelines such that anyone in the world can run this exercise.

This project is well into development. Please refer to playtest reports from v0.2.0 and v0.2.1 for more information as to the design philosophy of Endgame. I encourage you to read my initial blogpost proposing this project. For further context on the Endgame concept I also highly recommend Shut Up & Sit Down's coverage of Watch The Skies!, the most commercially successful Megagame to date.