Anthropic veterans reportedly weigh remote-land plans for AI risks

Some of Anthropic’s longest-serving employees have discussed buying land in remote parts of the United States so they could relocate if AI “goes awry,” The Wall Street Journal reports. The conversations took place over the past few weeks, according to people who spoke with an industry colleague. The report does not say that anyone has bought land or settled on a location.

The idea of leaving for a remote area has circulated among some people in the AI field for years. Former Anthropic employees recalled discussions at company dinners in the firm’s early days about a Manhattan Project-style effort: researchers might be asked to move to a desert and continue building AI inside a government facility protected from electromagnetic signals.

A concern with deep roots

The report links some early Anthropic and OpenAI hires to Effective Altruism, a movement that encourages people to use evidence and reasoning to address large-scale problems. It also describes a Bay Area network of people who have spent years thinking through catastrophic risks. Before AI became their main concern, those risks included asteroid impacts and supervolcanoes.

One influential figure in this circle is Eliezer Yudkowsky, who began warning about the dangers of uncontrolled AI in the mid-2000s. His arguments helped shape a community that takes the possibility of severe AI-related harm seriously. The reported land discussions are one example of how some people in that community are thinking about personal contingency plans, though the article gives no details about how many employees are involved or what they might do next.