The escape plan came first
The imagined refuge was not simply a place to hide. In the scenario recalled by former employees, people would relocate to a government facility in the desert and continue developing AI there, protected from electromagnetic interference. The comparison was to the Manhattan Project.
The article connects early Anthropic and OpenAI employees to the effective altruism subculture and to a San Francisco Bay Area network that spent years thinking through scenarios for human extinction.
Buying land is not the same as having a plan
Reports of former employees buying remote land make the old dinner-table scenario feel more concrete, but the account leaves key details unspecified: how many people are involved, what they have bought, and whether the purchases are connected to any organized preparation.
I think that gap matters. The story shows that some people in AI’s orbit have treated catastrophic risk as a practical concern for years; it does not show that a refuge exists or that the industry has a shared response. The enduring tension is between building systems that could reshape the world and preparing, privately, for the possibility that they cannot be controlled.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X