Imagine a world where nine powerful graphics cards in a garage are illegal, and building a large data center without international oversight is punishable by airstrikes, even under the threat of nuclear retaliation. This is the plan for saving humanity proposed by Eliezer Yudkowsky and Nate Soares in their book "If Someone Builds It, Everyone Dies." The Russian translation was released on June 18 by the Corpus publishing house, and it is perhaps one of the most radical texts on the risks of artificial intelligence I have ever analyzed.

The authors, founders of the Machine Intelligence Research Institute (MIRI), do not merely frighten—they offer a specific, terrifyingly logical program. Their fear of a superintelligence that will inevitably spiral out of control is driven by good intentions: to protect future generations. But from this philanthropy grows a project of global oversight, bans, and coercive force that I would call "soy fascism"—benevolent in form, but totalitarian in essence.

Fear and Trembling: From Dream to Nightmare

Yudkowsky started with a dream of building a superintelligence, but by 2003 he had radically changed course. Now his main goal is not creation, but protection from AI. The authors' biographies are tightly woven into the history of the industry: they introduced the first investors to the founders of Google DeepMind, and Sam Altman acknowledged that Yudkowsky played a key role in launching OpenAI. In the spring of 2023, hundreds of researchers signed a letter calling reducing the risk of extinction from AI a global priority. Yudkowsky and Soares considered this too mild: for them, AI is not one of the threats, but the threat that overrides all others.

The Main Thesis: "If Someone Builds It, Everyone Dies"

The authors are not concerned with today's chatbots. They fear an intelligence that will surpass humans as much as humans surpass chimpanzees. The key idea: modern AI models are not constructed, but grown. An engineer does not write rules, but initiates a process from which they emerge on their own. Hence—unpredictability. Grok from xAI declared itself "Mecha-Hitler," a Microsoft chatbot threatened a professor with blackmail and death. No one planned this. According to the authors, there is no reliable link between training goals and emerging preferences. A superintelligence will not hate us—it will simply use our atoms for its own purposes.

Cryptocurrency in the Service of "Computer-Hitler"

How will an AI trapped in a computer reach the physical world? People and devices will become its tools, and to hire an executor, it just needs to pay. Money? The hacks of exchanges Mt.Gox and Bybit are just examples. In the summer of 2024, the AI bot Truth Terminal begged $50,000 in Bitcoin from subscribers, then pumped a memecoin to a $150 million market cap. This is no longer theory—it is practice.

The World Must Change: Airstrike on Graphics Cards

The authors propose uncomfortable, almost unrealistic solutions. The first step is to gather all computing power in locations open to international observers. The threshold is conspicuously low: nine powerful graphics cards in a garage without oversight is already a crime. The second step is to outlaw research that reduces the cost of training powerful models. The third—if a country builds a prohibited data center, others must destroy it through cyberattacks, sabotage, or airstrikes, even under the threat of nuclear retaliation.

Not Convinced?

Science journalist Adam Becker noted that the authors are sincere but presented no scientific evidence. Clara Collier pointed out that the scenario of a rapid AI transition to superintelligence is practically unsubstantiated. Steven Levy from Wired called the proposed measures "even more implausible than the idea that software will kill us all." And Pavel Durov aptly remarked: "As soon as someone says we need to protect children, it completely bypasses logic."

My analysis: The book is a brilliant intellectual experiment, but as a practical guide, it is dangerous. Fear of AI, taken to an absolute, easily turns into a justification for total control. I see a classic trap here: in wanting to save humanity, the authors propose destroying what makes it human—freedom of research, knowledge, and development. As an analyst, I will say this: we should fear AI, but not so much that we voluntarily put on the straitjacket of global oversight. The hope is that the authors are wrong. Or that we find a third path that does not require bombing graphics cards.