Longterm Wiki has not been actively maintained since May 2026. Some data is still updated occasionally, but please don't rely on this site being current or accurate.
Most widely used framework for game-theoretic reinforcement learning research. Open-source library by Google DeepMind (Marc Lanctot et al.). Supports extensive-form games, normal-form games, and multi-agent RL environments.
Research foundation with $15M funding (from Macroscopic Ventures) studying how AI systems can learn to cooperate and how to design environments incentivizing cooperation. Co-founded by Allan Dafoe (now Director of Frontier Safety and Governance at Google DeepMind) and others. Published foundational "Open Problems in Cooperative AI" (2020). Key focus areas include multi-agent cooperation, preventing destructive AI competition, and human-AI coordination.
Researcher at OpenAI (formerly Meta AI). Built CICERO — AI achieving human-level Diplomacy (published in Science 2022), requiring negotiation, alliance formation, and multi-party coordination. Also built Pluribus (6-player poker AI).
Professor at MIT. Proved computing Nash equilibria is PPAD-complete, fundamentally resolving a question open since Nash's 1951 theorem. Nevanlinna Prize (2018). ACM Fellow (2022). Also contributed to optimal multi-item auction design.
Meta AI agent achieving human-level performance in Diplomacy — a game requiring negotiation, alliance formation, and multi-party coordination. Published in Science (2022). Demonstrated AI can learn complex multi-party coordination with humans.
Cooperative AI FoundationOrganizationCooperative AI FoundationResearch foundation with $15M funding (from Macroscopic Ventures) studying how AI systems can learn to cooperate and how to design environments incentivizing cooperation. Co-founded by Allan Dafoe ...
Other
Noam BrownPersonNoam BrownResearcher at OpenAI (formerly Meta AI). Built CICERO — AI achieving human-level Diplomacy (published in Science 2022), requiring negotiation, alliance formation, and multi-party coordination. Also...Constantinos DaskalakisPersonConstantinos DaskalakisProfessor at MIT. Proved computing Nash equilibria is PPAD-complete, fundamentally resolving a question open since Nash's 1951 theorem. Nevanlinna Prize (2018). ACM Fellow (2022). Also contributed ...
Analysis
CICEROProjectCICEROMeta AI agent achieving human-level performance in Diplomacy — a game requiring negotiation, alliance formation, and multi-party coordination. Published in Science (2022). Demonstrated AI can learn...AI Safety Culture Equilibrium ModelAnalysisAI Safety Culture Equilibrium ModelGame-theoretic model identifying three equilibria for AI lab safety culture: racing-dominant (current state, S=0.25), safety-competitive (S>0.6), and regulation-imposed (S=0.15-0.25). Key finding: ...Quality: 65/100Racing Dynamics Impact ModelAnalysisRacing Dynamics Impact ModelThis model quantifies how competitive pressure between AI labs reduces safety investment by 30-60% compared to coordinated scenarios and increases alignment failure probability by 2-5x through pris...Quality: 61/100International AI Coordination Game ModelAnalysisInternational AI Coordination Game ModelGame-theoretic analysis demonstrating that US-China AI coordination defaults to mutual defection (racing) because defection dominates when cooperation probability falls below 50%, with current esti...Quality: 59/100Racing Dynamics Game Theory Model
Concepts
Autonomous Cooperative AgentsConceptAutonomous Cooperative AgentsExamines the emerging category of AI agents that cooperate, negotiate, and make commitments on behalf of humans. Covers the delegation alignment problem (how an agent learns cooperative preferences...Quality: 50/100Pause / MoratoriumConceptPause / MoratoriumComprehensive analysis of pause/moratorium proposals finding they would provide very high safety benefits if implemented (buying time for safety research to close the growing capability-safety gap)...Quality: 72/100
Approaches
Open Source AI SafetyApproachOpen Source AI SafetyComprehensive analysis showing open-source AI poses irreversible safety risks (fine-tuning removes safeguards with just 200 examples) while providing research access and reducing concentration—with...Quality: 62/100
Risks
AI ProliferationRiskAI ProliferationAI proliferation accelerated dramatically as the capability gap narrowed from 18 to 6 months (2022-2024), with open-source models like DeepSeek R1 now matching frontier performance. US export contr...Quality: 60/100AI Development Racing DynamicsRiskAI Development Racing DynamicsRacing dynamics analysis shows competitive pressure has shortened safety evaluation timelines by 40-60% since ChatGPT's launch, with commercial labs reducing safety work from 12 weeks to 4-6 weeks....Quality: 72/100
Key Debates
Open vs Closed Source AICruxOpen vs Closed Source AIComprehensive analysis of open vs closed source AI debate, documenting that open model performance gap narrowed from 8% to 1.7% in 2024, with 1.2B+ Llama downloads by April 2025 and DeepSeek R1 dem...Quality: 60/100AI Safety Solution CruxesCruxAI Safety Solution CruxesA comprehensive structured mapping of AI safety solution uncertainties across technical, alignment, governance, and agentic domains, using probability-weighted crux frameworks with specific estimat...Quality: 65/100