Anthropic veterans reportedly looking to buy remote land in case AI goes awry

Some of Anthropic's longest-serving employees have discussed buying land in remote parts of the United States where they could relocate if artificial intelligence development goes badly wrong, The Wall Street Journal reported on September 26.

According to the report, the employees told an industry colleague in recent weeks that they are considering such a purchase as a concrete contingency plan. They are described as "some of its earliest employees."

A Manhattan Project scenario, discussed for years

The idea of fleeing AI is not new in this group. Former employees recall that at company dinners during Anthropic's early days, colleagues discussed a Manhattan Project-style scenario: they might be asked to move to the desert and keep building AI inside an electromagnetically shielded government facility.

Over the years, people in the same circles have also floated buying remote islands, stockpiling iodine pills or retreating to shielded bunkers in the desert, the Journal writes.

Roots in the effective altruism network

The report traces how closely the earliest hires at Anthropic and OpenAI are tied to the effective altruism (EA) subculture. Some of them come from a Bay Area network of "doomers" who have spent years gaming out doomsday scenarios. Eliezer Yudkowsky, who started warning about runaway AI in the mid-2000s, became the intellectual godfather of that subculture; before AI topped the list of threats, the group focused on catastrophic risks such as asteroid impacts and supervolcanoes.

Much of this activity centers on a co-working space in Berkeley called Constellation, where AI safety researchers discuss the latest risks over monthly dinners; Anthropic staff have their own office space there, people close to the community say. Employees also trade doomsday scenarios on a private Slack channel and stick a motto on their laptops: "Things will never be chill again." Some of these projects have drawn staffers and funding from an EA-aligned network that included Sam Bankman-Fried, the former FTX chief now serving a 25-year sentence.

Why it matters

The people making these contingency plans are among the people building the very technology they fear. Both OpenAI and Anthropic were founded on the principle of staving off AI dangers, and the safety community's influence has been particularly strong at Anthropic, which the Journal describes as being on the cusp of a $2 trillion public offering.

An Anthropic spokesman said the company has more than 3,500 employees who hold a wide variety of viewpoints. The report lands amid renewed public debate about AI risk: the Journal notes that this month's resignation of Anthropic researcher Jacob Coxon drew attention to the idea that runaway AI development could end humanity.