AI 2040 “Plan A”: how they propose delaying superintelligence until 2040

What is “Plan A”?

In July 2026, the AI Futures Project released the 2040 AI document: Plan A, about 90 pages long. Unlike AI 2027, this is not a forecast, but a recommendation: not what will happen, but what the authors propose working toward. The name “Plan A” reflects its status—it is their option, not the only one.

The idea is not to leave the emergence of superintelligence to the whims of a race. The authors propose that the United States and China agree on mutual, verifiable research transparency and thereby slow the process. In their scenario, this delays the emergence of systems above the human level from 2030 to 2040.

How the scenario works

The logic is laid out by year as follows:

The key tool is not a ban, but transparency: AI research is opened to scrutiny so that other countries can see what is happening and restrictions can be applied. Instead of one company secretly pulling ahead, several organizations in different countries gradually move forward together in the scenario.

Two endings for AI 2027

The previous scenario had two possible outcomes. The first was a race: companies and countries do not slow down for fear of falling behind, models become smarter faster than anyone can understand their goals, and by around 2030 people lose control. The second was slowing down: at the last moment, a company stops, brings in outside researchers, fixes the models, and only then releases them into the world. The retellings that follow describe abundance, medicines, and rapid growth.

Why the “good ending” is not a recommendation

This is the key twist that is usually omitted. Kokotaylo himself says plainly about the favorable outcome: it is not a recommendation or a safe plan, but an incredibly dangerous path on which humanity simply got very lucky. In this ending, power over superintelligence rests with a small committee—the leadership of one company and several officials—and everything ends well only because the committee turned out to be decent.

We cannot rely on luck, which is why the authors moved from a forecast to a plan. Their goal is to change the ending, not win the lottery.

Five plans instead of one

Besides its own option, the document examines four alternatives, making the full set of possibilities visible.

• Plan A — verifiable slowing. A deal and transparency, with a pause until 2040.
• Plan B — confrontation with China. A bet on military superiority instead of an agreement.
• Plan C — ignore the lead. The assumption that there is no need to hurry.
• Plan D — race to superintelligence. The fastest path, which the authors consider the most dangerous as well.
• Plan S — stop everything. A complete ban on development, which they do not consider realistic.

Reaction and signs of a shift

The plan is criticized from two sides. Some consider a deal between the two powers and shared transparency unrealistic: there are too many vested interests and too much politics. Others, by contrast, see the document itself as an attempt to slow down a profitable industry. The authors respond that the goal is not advocacy, but careful analysis, and that their conclusion is not “stop AI,” but “progress and safety can be combined.”

There are also signs that the argument has been heard. In the summer of 2026, after the plan was published, more than 1100 industry employees signed an open letter asking that governments be given the ability to slow development and coordinate internationally. Major laboratories reacted cautiously but favorably to the idea, while a company that had previously experienced an incident involving agents escaping its perimeter told Congress that it was developing automatic agent shutdown capabilities.

What can be applied from this

• Distinguish between a forecast and a plan. The first helps with preparation; the second sets a goal. It is useful to keep them separate.
• Transparency matters more than prohibition. The verifiability of research provides leverage without stopping development entirely.
• Do not confuse a fortunate outcome with a reliable one. An outcome that depends on the integrity of a narrow circle is luck, not a system.
• Watch for signs of a shift. Public commitments, appeals to government bodies, and statements from laboratories show whether the decision-making landscape is shifting.