Document Intermediate Scenario 150 min read

The AI 2040 Scenario (Plan A — The Deal)

AI Futures Project (Kokotajlo, Larsen, Lifland, Dean, Halstead, Greenblatt) · AI Futures Project · 2026-07

A nonprofit forecasting group led by Daniel Kokotajlo, the former OpenAI governance researcher who publicly gave up his equity rather than sign a non-disparagement agreement, and whose earlier scenario AI 2027 became one of the most-read AI documents of its year. Co-authors include the forecaster Eli Lifland and Redwood Research's Ryan Greenblatt.

Read the original at AI Futures Project ↗ Open the reading companion

The team whose AI 2027 scenario predicted that racing to superhuman AI ends in extinction or an irreversible concentration of power here writes the other branch: their positive plan. In this story, the US and China agree in 2029 not to race; AI research is made public; frontier development pauses at top-human-expert level in 2035 under a regime of mutually verified compute — and only in 2040, with alignment far better understood, does the world deliberately step up to superintelligence. Year by year, the scenario tracks employment, politics, and the temptations to defect.

Read it not as a prediction but as a negotiating position: this is what the people most alarmed about AI think a good outcome would require, spelled out concretely enough to argue with. The scenario form makes assumptions visible in a way position papers never do — and read against AI as Normal Technology, it brackets the live debate. The core narrative reads faster than the page count suggests; extensive appendices are optional.