The scenario of human extinction due to artificial intelligence is not science fiction, but a very real risk that experts estimate at 10–30%. However, we now have "Plan A" — a roadmap that, according to its authors, can not only slow down the race but also maintain human control over superintelligence.

The document AI 2040: Plan A was presented by Daniel Kokotajlo, a former OpenAI researcher who left the company in April 2024 due to fundamental disagreements over safety issues. In 2025, he founded the AI Futures Project, which prepared this ambitious scenario.

How "Plan A" Should Stop the Race

The key idea is an international agreement between the US and China, which should be concluded in 2029. Without it, as the authors predict, automation of AI development will occur as early as 2030, leading to an uncontrollable explosive growth.

Instead, countries commit to developing neural networks only to the level of the best human experts. By 2035, a full pause in development is introduced to preserve human control. And only in 2040 is the "emergency brake" lifted, and AI reaches the level of superintelligence — but already in a safe, verifiable environment.

The plan is based on four principles: buying time for safety research, full transparency of development, decentralization of AI among companies and countries, and maintaining reversibility of all processes.

Verification and Mutual Destruction

To ensure mutual trust, the authors propose using physical verification. Large data centers are visible from space — they cannot be hidden. The first step: countries publicly declare purchases of AI chips. Then a temporary pause on new training runs is introduced, monitored by sensors at the facilities.

The most radical element is "mutually assured destruction of computing power." Following the logic of nuclear deterrence, it is proposed to build new Chinese data centers in Canada, and US facilities in Mongolia. In the event of a conflict, the host country would try to seize the capacity, and the owner would destroy it to prevent it from falling into enemy hands.

Economics and Social Consequences

The calculations are impressive: global computing power will grow from 20 million H100-equivalents in 2026 to 60 billion by 2034. Real US GDP growth in certain periods of the 2030s will reach 50% per year — compared to the usual 3%.

However, automation will collapse employment: from 62% in 2027 to 12% by 2040. To compensate, "civilian dividends" are proposed — payments to every adult American from state revenues from licenses for computing and robots. Forecasts: $45,000 per person in 2032, $1 million by 2035, and $10 million by 2039.

Alternative Scenarios

The authors have provided four backup paths. Plan B — the US creates a coalition and pressures China, including cyberattacks. Plan C — an attempt to negotiate, but under business pressure the pause is quickly lifted, leading to oligarchy. Plan D — minimal regulation and a race with the risk of war. Plan S — a complete indefinite stop, which would most likely collapse, returning the race in less controlled conditions.

My expertise: "Plan A" sounds like science fiction, but its logic is flawless. The problem is that it requires an unprecedented level of global trust and coordination — precisely what we currently lack. In a world where geopolitical tension is rising and AI profits are measured in trillions, the chances of a voluntary pause until 2040 are close to zero. However, the very framing of the question — "how to survive, not how to accelerate" — is already a significant step forward for the industry.