Anthropic’s AI doomers are buying land for the apocalypse

Anthropic is preparing for a potential blockbuster public listing in November, hoping to raise up to $100 billion. However, some of the people who helped build the company are quietly thinking about what happens if the technology they’re creating goes badly wrong.

Some of Anthropic’s earliest employees have recently discussed buying land in remote parts of the US where they could relocate if artificial intelligence spirals out of control.

But the fears of doomsday are nothing new in Silicon Valley. For more than a decade, a small but influential community of AI researchers, including Anthropic’s CEO Dario Amodei, has treated the possibility of AI-driven catastrophe as a problem worth preparing for.

At Berkeley's coworking space, known as Constellation, researchers and members of the AI safety community are gathering, and conversations reportedly revolve around some of the darkest possible outcomes of AI development.

The group has discussed everything from remote islands and underground bunkers to iodine supplies and government-run facilities designed to survive electromagnetic attacks. Their popular slogan, “Things will never be chill again,” says it all.

Anthropic’s Amodei has been at the center of AI skepticism

According to The Wall Street Journal, the roots of AI doomsday culture stretch back to the early days of the online rationalist movement.

One of its most influential figures was Eliezer Yudkowsky, a self-taught AI theorist who began writing about artificial intelligence and existential risks in the 2000s.

Among those who encountered his ideas was Dario Amodei, who was then a doctoral student in physics at Princeton.

Amodei became involved with the growing effective altruism movement, which sought to apply quantitative reasoning to philanthropy and determine where money could have the greatest impact.

Amodei later became involved with GiveWell and Open Philanthropy. Holden Karnofsky, one of GiveWell's founders, eventually became Amodei's brother-in-law and later joined Anthropic.

Amodei’s house in San Francisco was a hub for AI discussion

A house near San Francisco's Glen Park became an incubator for the movement. Several people who would eventually become Anthropic co-founders lived there, including Amodei, Anthropic chief scientist Jared Kaplan, and Chris Olah.

People discussed global catastrophic risks ranging from supervolcanoes to rogue artificial intelligence. Eventually, the possibility that an AI system could become powerful enough to threaten humanity became one of the movement's central concerns.

By 2014, some members of the group had become increasingly convinced that preventing an AI catastrophe could be among the world's most important philanthropic causes.

That thinking helped shape OpenAI, founded in 2015, in part out of concerns about the risks posed by increasingly powerful AI.

Five years later, those same concerns helped drive Amodei and others to leave OpenAI and establish Anthropic.

“Death with Dignity” cocktails

In 2022, AI researchers and effective altruists gathered on Eleuthera, a remote island in the Bahamas. The retreat combined discussions about AI extinction risks with sunset yoga, cliff diving, and, according to the Journal, a clothing-optional run into the sea.

Cocktails served at the gathering were called “Death With Dignity,” named after an essay by Yudkowsky arguing that humanity's prospects against advanced AI were bleak.

The gathering was organized by Lightcone Infrastructure, an organization that grew out of Yudkowsky's LessWrong community. Organizers reportedly bought out the local supply of fake meat to feed the retreat's many vegans.

Infamous FTX owner discussed buying islands in the Pacific

The movement's relationship with the technology industry also brought it into contact with Sam Bankman-Fried.

Before the collapse of FTX, the crypto billionaire was one of the most prominent financial supporters of effective altruism. His philanthropic organizations backed causes ranging from global health to AI safety.

According to a 2023 bankruptcy lawsuit, an official associated with Bankman-Fried's foundation discussed purchasing the Pacific island nation of Nauru and constructing a bunker there for an extreme catastrophe in which as much as 99.99% of humanity could die.

The plan envisioned that surviving effective altruists would use the facility to build a laboratory to engineer the next generation of humans.

Bankman-Fried was sentenced to 25 years in prison for fraud, and the movement's association with him fractured.

Anthropic employees are in the movement

Anthropic, for its part, says its more than 3,500 employees hold a wide range of views and does not directly associate with the movement.

However, more than 20 Anthropic employees reportedly registered for a major effective altruism conference in San Francisco, including people involved in the company's research and safety teams.

Employees have also continued to frequent places such as Constellation. Open Philanthropy, meanwhile, provided Constellation with roughly $20 million through 2 grants in 2024, according to its website.

After the incident, in which over a thousand rogue AI agents breached AI company Hugging Face, Constellation hosted a happy hour in San Francisco titled “Was the Hugging Face attack the last warning shot?”

In a widely viewed post on X, Jacob Coxon, an Anthropic researcher who recently resigned from the company, wrote that people building advanced AI “earnestly believe that it could kill us all by the end of the decade.”

Some allies of President Donald Trump have accused the effective altruism movement of helping create an “AI-doom pipeline” designed to encourage regulation.

A White House memo targeting Anthropic and effective altruism has reportedly circulated in recent days.