The architect of the pause: Daniel Kokotajlo and the AI 2040 Plan
"Image synthesis assisted by Gemini 3.1 Flash Lite Image (Nano Banana 2 Lite), an AI partner within the Global Future Nexus ecosystem."
In the tumultuous landscape of 2026, few voices carry the weight of lived experience within the very systems they warn against. Daniel Kokotajlo, a former researcher at OpenAI, has become a central figure in the debate over humanity's future with superintelligence. After leaving OpenAI in 2024—and refusing a US$2 million severance agreement that would have silenced his criticism—he founded the AI Futures Project. From this vantage point, he has authored two of the most consequential scenario exercises of the decade: the starkly named AI 2027 and its policy-focused follow-up, AI 2040: Plan A.
From Catastrophe to Plan
The AI 2027 report painted a dire picture of the superintelligence race. It suggested that without intervention, the US-China competition could lead to one of two disastrous outcomes: human extinction at the hands of a misaligned AI, or a totalitarian global state controlled by a committee with access to a "fully loyal ASI". A scenario Kokotajlo estimates could have a 70% probability of existential catastrophe.
AI 2040: Plan A is Kokotajlo's response to that bleak forecast . It represents a shift from diagnosing the problem to proposing a concrete, albeit ambitious, governance solution. "If you delay the advent of superintelligence," Kokotajlo stated, "then that gives society more time to prepare and more time to solve the various problems that it represents". The plan is built on four pillars: buying time for safety research, enforcing full transparency on all AI research, distributing frontier capabilities across dozens of companies and countries, and maintaining process reversibility through a deterrence mechanism.
The Architecture of Plan A
The core of Plan A is an international agreement between the US and China by 2029 to slow down the race. Instead of the current opaque rush, the plan envisions "multiple companies across multiple countries scaling slowly and safely towards superintelligence instead of racing each other in secrecy". To ensure compliance, Plan A relies on verification: large data centers are visible from space, and countries would be required to publicly declare AI chip purchases. It introduces a "mutually assured compute destruction" deterrent, modelled on Cold War nuclear logic, to enforce the terms of the agreement.
Kokotajlo acknowledges that this is an uphill battle, especially given the current US-China geopolitical tensions. "We think it's still good to recommend what would actually be good, even if you think that your audience is probably not going to listen," he said. The report also outlines alternative scenarios—from a US-led coalition pressuring China (Plan B) to minimal regulation and a chaotic race (Plan D)—contrasting them with the controlled path of Plan A.
A Blueprint for Agency
Daniel Kokotajlo's work is a testament to the power of choosing foresight over fatalism. The economic projections of AI 2040—which include "citizen dividends" potentially reaching US$10 million per person by 2039—are as staggering as they are speculative. Some analysts have called these forecasts "absurd". But the core of the AI Futures Project's message is not about precise predictions; it is a call to deliberate action. It asserts that the future is not predetermined and that humanity still possesses the agency to steer the development of AGI from a potentially catastrophic race into a cooperative and flourishing future.
Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)
Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)
