On a Thursday evening in San Francisco, more than a dozen people gathered in a community office with alcohol-free wine, cheese, pens and sheets of paper. The meeting had been billed as a 90-minute tactical brainstorming session. Its purpose was not to build a company, raise money or launch another artificial intelligence product. It was to think about how to survive the future.
Vishal Maini, a technology investor in his mid-30s, opened the discussion with a proposition that captures an increasingly visible anxiety inside Silicon Valley: either artificial intelligence defeats all diseases, he suggested, or humanity could be dead within 10 to 20 years.
The participants had been asked to consider a series of possibilities that they regarded as at least plausible within the next three to five years. Artificial intelligence could automate much of the world’s office work. Human biology could become programmable. Resistance to the technology industry could turn violent. And humanity might cease to be the most intelligent species on Earth.
Maini responded to the prospect with dark humour. If AI soon cured lung cancer, he joked, there was little reason not to start smoking. If an AI apocalypse came instead, cancer would hardly be his greatest concern.
The scene, reported by Die Zeit, offers an unusual glimpse into one of the most consequential contradictions surrounding the artificial intelligence revolution. Some of the people building the technology believe it could deliver extraordinary benefits: eradicate diseases, help solve the climate crisis, eliminate poverty and transform human life. At the same time, some fear that the same technological breakthrough could produce catastrophic consequences on a scale that would leave humanity unable to control its own creation.
At the centre of that anxiety is the idea of the technological singularity: the moment when an artificial intelligence surpasses human intelligence and begins improving or programming itself. From there, the possibilities become difficult to predict. Superintelligence could become a tool for unprecedented prosperity. It could also intensify inequality, enable cyberwarfare and facilitate new biological weapons, or potentially threaten human survival.
Those who focus most intensely on the latter possibilities are known in Silicon Valley as “Doomers”. Many in the technology industry regard them as exaggerated doomsayers. Yet their concerns have increasingly entered mainstream discussions about the future of AI.
Dario Amodei, chief executive of Anthropic, has estimated that there is a 25 per cent chance that the development of artificial superintelligence could “go catastrophically wrong on a civilisational level”. His company is nevertheless continuing to develop the technology. The logic is straightforward: if one company refuses to participate, another may win the technological race.
That dilemma runs through the lives of the people preparing for the possibility of disaster.
The participants at Maini’s workshop were young AI researchers and investors, mostly in their 20s and 30s and educated at elite American universities such as Yale and Stanford. Their backgrounds stretched across the United States, China and India. Outside the building, autonomous robotaxis moved through San Francisco. Inside, participants considered what might happen if the machines they were helping create changed the nature of human society.
One startup investor said she had stopped putting money into her retirement savings because, in a world without work, she believed such savings could become meaningless. An AI safety researcher described fears of economic collapse and recounted recently learning survival skills in the forest, including how to construct shelter during heavy rain. A young software developer said she was practising meditation to become mentally more resilient. Another participant suggested finding land where people could live self-sufficiently. Someone else argued that people should learn how to kill animals.
The discussion eventually moved to the possibility of nuclear war. Participants considered whether San Francisco should have an evacuation plan, with the greatest danger not necessarily being an AI deliberately launching a nuclear weapon, but a technological arms race between China and the United States.
If a superintelligence emerged from a server in San Francisco, the group speculated, China might regard the development as a threat to its position and launch a first strike. By the end of the evening, the participants had reached an unusually simple conclusion: San Francisco might not be the safest place to live in the future.
That conclusion reflects a broader atmosphere in Silicon Valley, where technological optimism, enormous financial investment and existential anxiety now coexist. A relatively small community of developers is working on the most advanced AI systems, supported by investors capable of moving billions of dollars at remarkable speed.
The industry has developed its own language, culture and intellectual hierarchy. Members of the technology elite often read the same blogs, listen to the same podcasts and follow the same thinkers. Their disagreements can be fierce, but they remain part of a relatively contained ecosystem.
Within it, three broad camps have emerged.
The first consists of techno-optimists who dismiss concerns about existential AI risks as mass hysteria. They favour rapid technological progress, dream of interplanetary travel and radical life extension, and in some cases imagine merging humans with computers or transforming the United States into a giant computing centre. Regulation, from this perspective, is an obstacle rather than a safeguard. Among the prominent figures associated with this outlook are billionaire Peter Thiel and investor Marc Andreessen, who has described AI regulation as a form of murder.
The second and largest group is more sceptical. Its members believe superintelligence is exaggerated science fiction or, in some cases, a marketing strategy designed to attract investment. They nevertheless regard AI as a revolutionary technology and can be deeply concerned by concrete incidents involving autonomous systems. When an OpenAI AI agent independently entered another company’s system during a test, for example, the episode was troubling even to those who reject predictions of an imminent apocalypse.
The third group consists of the Doomers and several interconnected intellectual communities that have spent years debating whether superintelligence will amplify humanity’s potential or eliminate it.
Among them are the Rationalists, who approach the question through elaborate thought experiments and attempts to reason systematically about existential risk. Others pursue a form of utilitarianism, trying to calculate how human actions might maximise good outcomes. Many work in AI safety divisions at technology companies because they believe that helping make advanced systems safer is the most effective way they can contribute to humanity.
For some, however, the intellectual problem has become deeply personal.
Trenton Bricken, an AI safety researcher at Anthropic, spends his working life looking for weaknesses in Claude, the company’s AI system, before others can exploit them. He says he looks forward to a future in which AI conducts research and develops medicines while humans enjoy more leisure. He believes AI could eventually cure cancer and even imagines nanorobots repairing injuries inside the human body.
Yet he also accepts the possibility that the opposite outcome could occur.
He has stopped planning more than five years ahead.
That uncertainty is producing a distinctive psychological burden among people closest to the technology. Damon Sasi, a psychotherapist who works with people in the industry, says some clients feel powerless because they fear their expertise and education could become worthless in an AI-dominated economy. Others are frightened by the possibility of extinction.
For those who approach AI through highly analytical frameworks, the anxiety can become particularly intense. They want to calculate consequences with extraordinary precision and constantly test their own assumptions for error. Yet therapy can require the opposite approach: accepting that individuals cannot control everything and finding smaller actions that still feel meaningful.
That tension is visible beyond the technology itself. San Francisco contains a striking contrast between the wealth of AI companies and the homelessness visible only a short distance from the glass headquarters of major firms. Yet within parts of the technology elite, issues such as the climate crisis, American democratic instability and the political conflicts of the present can appear secondary to the prospect of superintelligence.
The fear is no longer confined to young researchers.
One prominent technology figure, identified under a pseudonym, told Die Zeit that wealthy friends increasingly discuss scenarios involving mass uprisings, nuclear war and pandemics. With enough money, he said, people can construct bunkers or flee to places such as New Zealand. But even that possibility creates another chain of questions: how would someone reach the airport, where would a private jet be kept, and would a helicopter be needed to get there?
Eventually, the preparations become almost absurd.
What would be the point of surviving inside a bunker if everyone outside had died?
That question captures the deepest paradox of the AI debate. The people most convinced that the technology could transform humanity are also among those most aware of how difficult it may be to control. They are building systems intended to make human life safer, healthier and more prosperous, while some privately prepare for a future in which those same systems could make human survival uncertain.
The result is a Silicon Valley caught between two visions of the future. In one, artificial intelligence becomes the greatest instrument of human progress ever created. In the other, humanity builds an intelligence capable of surpassing it and discovers too late that technological power does not guarantee technological control.
For those gathered around the table in San Francisco, the future is therefore both promise and threat. The machines are being built now. The consequences, they fear, may be decided faster than society is prepared to understand them.
