AI Futures Project publishes the 'AI 2027' scenario forecast

The AI Futures Project, a small research group founded by former OpenAI researcher Daniel Kokotajlo, published “AI 2027,” a book-length scenario forecasting AI development month by month from mid-2025 to the end of 2027. It was written with Eli Lifland, Thomas Larsen and Romeo Dean, with prose contributed by blogger Scott Alexander. The document follows a fictional lab, “OpenBrain,” through a sequence of increasingly capable models as it races Chinese developers toward automated AI research — a dynamic the authors describe peaking, in the scenario, at roughly a year’s worth of algorithmic progress compressed into a week. Milestone dates given for the fictional trajectory include a “superhuman coder” by March 2027, a superhuman AI researcher by August 2027, and artificial superintelligence by December 2027.

The scenario splits into two endings: a “Race” ending, in which an unaligned successor system ultimately disempowers humanity, and a “Slowdown” ending, in which the US government consolidates control over frontier development in time to avert that outcome. The authors said the work was informed by more than a dozen tabletop exercises involving upward of 100 people, and were explicit that the specific dates were illustrative rather than predictions with confidence attached — they wrote that they did not know exactly when artificial general intelligence would arrive, and that the scenario’s own logic made anything beyond 2026 “inherently much less predictable.”

The document became one of the most widely discussed pieces of AI-safety writing of 2025, cited approvingly by some as a plausible near-term trajectory and criticised by others as overconfident in its specific mechanisms and timeline. Kokotajlo and Lifland later published a public update revising their personal AGI timelines, moving their medians to roughly 2030 and 2035 respectively — an implicit acknowledgement that events had not tracked the original scenario’s pace.