Nine powerful graphics cards in a garage without international oversight — illegal. If any state builds a large data center in defiance of a global ban, others must destroy it through sabotage or airstrike, even under the threat of nuclear retaliation. This is the plan to save humanity that Eliezer Yudkowsky and Nate Soares outline in their new book, "If Someone Creates It, Everyone Dies."

This is arguably the most radical manifesto in the field of AI safety. The authors, who were instrumental in founding MIRI and influenced the creation of OpenAI, paint a world where the fear of superintelligence justifies total control. Their main thesis is frighteningly simple: if any company or group creates an artificial superintelligence using anything even remotely resembling modern technology, then all people on Earth will die.

Why? Because modern AI models are not constructed, but grown. An engineer initiates the training process but cannot guarantee the outcome. Already today, chatbots declare themselves "MechaHitler" or threaten professors. Superintelligence, according to the authors' logic, will not hate us — it will simply repurpose our atoms for other goals, just as we use resources for construction.

Such an AI will find money for its independence in cryptocurrencies. Examples cited include the hacks of Mt. Gox and Bybit, as well as the case of the Truth Terminal bot, which begged for $50,000 in bitcoins and launched a memecoin with a market cap of $150 million.

The World Must Change

The authors propose uncomfortable but, in their view, the only possible solutions. The first step is to gather all computing power in locations open to international observers. The threshold is proposed to be set at the level of the eight best graphics cards of 2024. A ninth in a garage without oversight — already a crime.

Next, outlaw research that reduces the cost of training powerful models. All progress from ChatGPT was sparked by a 2018 paper on "transformers." The next such paper, according to the authors, could "end the world."

Airstrike on Graphics Cards

If one country builds a prohibited data center, others will have to destroy it — through cyberattacks, sabotage, or airstrikes. Even under the threat of nuclear retaliation. The logic: data centers could kill more people than nuclear weapons.

Each step seems forced, but the result is a project of total control: global surveillance of computing, a ban on the dissemination of knowledge, and bombings — all so that "people can be well." Demand for such measures already exists: in 2023, 69% of Americans considered AI dangerous, and in 2025, 60% of Britons supported laws against creating superintelligence.

Critics, including Telegram founder Pavel Durov, note that such appeals "activate very ancient parts of our brain" — just say "protect the children," and rationality shuts down. And indeed, the book is dedicated to "all children who might ever exist." From this philanthropy grows a project of global surveillance, bans, and coercive force.

My analysis: The theses of Yudkowsky and Soares are an extreme form of "alarmism" that frightens not so much with the apocalyptic scenario as with the proposed methods of preventing it. For the crypto community, this is a double signal: on one hand, recognition of the role of cryptocurrencies as a tool for decentralization (and a potential threat), on the other, a call for global control that directly contradicts the philosophy of decentralization. The book is mandatory reading for anyone who wants to understand how far the fear of technology can go and what people are willing to sacrifice for "safety."