About
Thomas Larsen is a researcher at the AI Futures Project, a US nonprofit that forecasts the development and societal impact of advanced AI. He co-wrote AI 2027 (April 2025) with Daniel Kokotajlo, Eli Lifland and Romeo Dean, a scenario of AI transforming the world by the end of 2027, and then started and led its successor, AI 2040: Plan A, the project's recommendation for what should happen instead. Plan A proposes a verifiable international slowdown that delays superintelligence to 2040. Before joining the project he founded the Center for AI Policy, an AI safety advocacy organization, and did AI safety research at the Machine Intelligence Research Institute.
Key Contributions
- Co-wrote AI 2027 (April 2025) with Daniel Kokotajlo, Eli Lifland and Romeo Dean, a scenario forecast of AI's development between 2025 and 2027
- Started and led AI 2040: Plan A, holding final editorial authority over a scenario written as a recommendation rather than a prediction
- Argues in Plan A for a verifiable international slowdown that buys a decade, delaying superintelligence to 2040 instead of around 2030
- Names loss of control, concentration of power and conflict over AI as the problems Plan A is built to solve, with transparency as a core principle
- Founded the Center for AI Policy, an AI safety advocacy organization, after doing AI safety research at the Machine Intelligence Research Institute
