AI Futures Project's Plan A recommends delaying superintelligence to 2040
A detailed 16-year blueprint to avoid AI catastrophe, co-authored by top alignment researchers.
The AI Futures Project, a collaborative group of alignment researchers including Daniel Kokotajlo, elifland, Thomas Larsen, and others, has released 'AI 2040: Plan A' — their most ambitious scenario yet. This is not a forecast but a prescriptive roadmap: it describes how governments in the US and China could cooperatively delay the advent of superintelligence from an expected 2030 to 2040. The project emphasizes granular, concrete steps over high-level abstractions like 'doom' or 'utopia', aiming for a plausible and actionable path forward.
Published July 9, 2026, the scenario is available at ai-2040.com with an optimized reading experience on desktop (audio and mobile versions exist but are less polished). The work follows their earlier 'AI 2027' scenario and represents a shift toward planning rather than pure forecasting. The team plans to engage with feedback and may produce further scenarios, though they acknowledge the enormous time commitment. Early criticism from AI safety researcher Chris Leong flags concerns about 'total research transparency' assumptions and potential first-mover disincentives.
- Co-authored by leading alignment researchers including Kokotajlo, elifland, and Greenblatt; aims to delay superintelligence to 2040
- Based on US-China government cooperation; assumes superintelligence would otherwise emerge in 2030
- Full interactive scenario with audio and mobile support available at ai-2040.com
Why It Matters
A concrete, detailed plan to navigate the most critical decade in AI safety, shifting from prediction to proactive governance.