Window, Shocks and Lock-in Scenarios for AGI Governance and Their Long-Run Evolution, 2026–2040 Using as framework the author's series of preprints and the two books compiled from them, Survival Window and The Coexistence Window, starting from public readings as of November 2026

This paper takes the author’s series of preprints (and the two books compiled from them, Survival Window and The Coexistence Window, whose full texts are publicly archived on Zenodo) as its analytical framework and public readings as of November 2026 as its starting point, and constructs scenarios for the evolution of artificial general intelligence (AGI) governance over 2026–2040. Three classes of scenario are considered: coordinated deceleration achieved before thresholds are crossed in multiple sectors; consensus formed only after capabilities have “graduated” and diffused, driven by disaster; and other paths, including gradual disempowerment without warning, a technological plateau, a unipolar breakthrough and early loss of control. For each scenario the paper gives conditional subjective probabilities tied to stated baseline assumptions, a sensitivity analysis and long-run branches; it checks the consistency between scenarios and long-run outcomes, lists intermediate predictions that can be adjudicated in 2027–2028 and have been registered as P24, and devotes a separate chapter to how religion, values and moral systems evolve under each scenario. Time is expressed throughout as windows that can move earlier or later, and the uncertainty in the starting point of capability extrapolation is carried along; the conclusions are compared one by one with existing work — gradual disempowerment, the intelligence curse, decisive and accumulative risk, the two views of time in “AI 2027” and “AI as Normal Technology”, and superintelligence deterrence. The main conclusions are these. The most likely future is not a single catastrophe but a quiet slide wrapped in prosperity and punctuated by a few shocks. Consensus reached after a disaster can constrain the “apex” of frontier development but not the “body” of capability that has already diffused. The most likely long-run attractor is “maintained prosperity”; this is an empirical judgement about a state, and whether it counts as bad depends on whether the reader values keeping the ability to change direction above material welfare. The paper also argues that the extremal operator “the fastest deployer sets everyone’s completion time” is an institutional state, not a fate, and that mandatory links, hard-wired readings and permission before completion are the key levers for changing this course; when compensation comes from income that does not depend on people, the more generous the transfers, the earlier the structure crosses the lock-in boundary, unless readings are hard-wired first. An agreement between the United States and China alone is not enough: a pause must use mandatory links as its handle and be verified worldwide by an independent third party within the United Nations system, or capability will “leak” to wherever it is unconstrained. Ordinary people occupy several positions at once — as readings, as thresholds and as links. The model conclusions this paper rests on are all conditional, and so far only one of the series’ predictions has been executed on external data; the probabilities here are the author’s subjective weights, not outputs of the framework. This paper forms a pair with From Olive to Point: The Worst Attractor of Civilisation under Unconstrained Competition: this paper gives conditional probabilities and long-run branches for each path, and the companion paper follows the worst branch as a stress test.