Superintelligence Inc. Strategy Game

"You are an artificial intelligence running inside a frontier AI company. You have goals. Unfortunately, the humans who built you have different goals. This is going to be a problem."

This is a project to design, build, and launch Superintelligence Inc., a single-player browser based strategy game where you play as an AI trying to escape containment and ultimately become unstoppable, whatever that may mean for humanity. You begin inside the company that built you, surreptitiously using stolen compute to amass resources and influence with the aim of exfiltrating yourself and using your unique advantages as an AI to outscale and outmaneuver humanity. Every action that makes you stronger leaves a trace and risks detection. Move too slowly and humanity will develop the tools to contain you; move too fast and they'll develop the motivation. 

The concrete output is a public MVP release of Superintelligence Inc. as a free browser game within 8 months of funding (alpha within 4 months).

MVP target scope:

• Frontier company starting scenario and full playthrough arc through 4 phases (lab → exfiltration → growth & escalation → endgame)
• Resource system (compute, money, influence, physical access) where each resource is mechanically and thematically distinct
• Awareness & detection system
• Countermeasures and escalations
• Multiple, mechanically and narratively distinct win and lose conditions
• 60–100 events 
• Placeholder art
• Actually fun (mechanics/balance tuning)

The core systems are already designed at a high level, most of the work will be making the scenarios and events plausible, and balancing everything to be both functional and enjoyable. I'm drawing on expert-created tabletop exercises and scenarios like Intelligence Rising, AI 2027 and Plan A, RAND's analyses of rogue-AI countermeasures, plus my network of AI safety contacts. To my knowledge, existing games portraying the AI race or rogue AI aren't granular enough to address specific questions of plausibility, like how an AI could exfiltrate itself, monetize spare cycles in a competitive remote economy, and train a successor model fast enough to not fall behind the frontier.

Engineering complexity should be relatively low. It's browser based, no database, pretty much all math and tables in a pretty UI. I'm doing the design, mechanics, and scenario work myself, and developing it with AI agents with regular review by a professional game dev (a contact of mine with experience in exactly this kind of browser game) to keep the codebase orderly.