Anthropic veterans reportedly weigh remote land as AI contingency plan

Some of Anthropic’s longest-serving employees are reportedly making concrete contingency plans in case AI spirals out of control. According to The Decoder, citing the Wall Street Journal, some early Anthropic employees told an industry colleague they are considering buying land in remote parts of the U.S. where they could relocate if AI “goes awry.”

The key takeaway: this is framed as a contingency idea among some veteran employees, not a company-wide plan. Readers should distinguish between reported individual planning and Anthropic’s official operations or policy positions.

The report connects the land-buying idea to Anthropic’s and OpenAI’s early ties with Effective Altruism and a Bay Area network of AI-risk-focused “doomers.” The Decoder says the group has spent years gaming out end-of-the-world scenarios, with AI becoming a central concern after earlier focus on risks such as asteroid impacts and supervolcanoes.

Former employees also recalled early Anthropic dinner discussions about a Manhattan Project-style scenario, where people might be asked to move to the desert and continue AI work inside an electromagnetically shielded government facility. That context helps explain why extreme contingency planning may be familiar inside some AI-safety circles.

The report highlights a broader tension in frontier AI: people building advanced systems may also be deeply concerned about worst-case outcomes. For readers tracking AI governance, the important questions are whether these private fears translate into public safety practices, policy proposals, or changes in how labs communicate risk.

Watch for further reporting that separates personal preparedness from institutional safeguards. Also look for how AI companies explain the gap between their safety concerns and their continued push to develop more capable models.