WSJ reports Anthropic employees buying remote land and prepping doomsday plans amid AI fears

Anthropic Employees Prepping for Doomsday Scenarios

Anthropic staff are buying remote land and training for catastrophic AI outcomes in private company Slack channels, according to a Wall Street Journal report. The company maintains its own office space within Berkeley's Constellation AI safety community, and the wider safety scene has discussed acquiring iodine pills, remote islands, and electromagnetically shielded desert bunkers as precautions against existential AI risk.

The practical logic of these preparations collides with a fundamental game-theory problem. If an AI safety believer constructs an elaborate, well-stocked bunker in a remote area without embedding themselves in the local community first, they become what preppers call "a loot drop"—a visible stash of supplies that attracts opportunistic raiders when crisis hits. The viable version requires building social fabric years in advance: shared community, mutual obligation, people who know and trust each other before the scenario unfolds.

The harder irony sits underneath. Jaan Tallinn, the Skype founder and primary funder of the AI doom movement, warns publicly about racing to build technology that could end humanity. Yet even someone with Tallinn's resources and conviction faces the same constraint: you cannot simply buy your way to safety in a true civilization-ending scenario. You have to build trust and integration into a community that will protect you, or your bunker becomes an obvious target the moment scarcity kicks in.

This instantiates a real answer to why doomsayers do not make more aggressive financial bets against their own beliefs—not because bioterrorism or collapse scenarios are implausible, but because the actual mechanics of survival in those scenarios are social and local, not individual and technological.