According to a report by The Wall Street Journal on September 26, to prevent artificial intelligence from going completely out of control, senior employees at Anthropic Digital are quietly developing specific contingency plans. In recent weeks, some early employees have revealed to industry peers that they are planning to purchase land and build hidden shelters in remote and desolate areas of the United States, in case of a scenario where "AI goes out of control and turns against humanity," allowing their families to evacuate and seek refuge.

The Idea of "Fleeing AI" Has Been Around for a While, Internal Simulations of Apocalyptic Scenarios

For these individuals, the idea of "fleeing AI" is not new. Early former employees recall that even during the initial company dinners when Anthropic was founded, people had passionately simulated a "Manhattan Project-style" hypothetical scenario: in the future, they might be urgently called upon by the government to work deep in the desolate desert, in a secret military base equipped with full electromagnetic pulse shielding, where they would continue researching cutting-edge AI behind closed doors.

The report states that Anthropic employees have discussed various apocalyptic destruction scenarios and explored response plans in internal private Slack channels. They even made a motto into a sticker that they placed on the covers of their notebooks as a guiding principle: "Nothing ever goes back to normal."

Company Response: Employees Have Diverse Views

In response to this cultural phenomenon, an Anthropic spokesperson said that the company currently has more than 3,500 employees, and internal discussions about technology reflect extremely diverse views and positions. The report highlights that even those working at the core of AI development have made the worst-case preparations as AI capabilities rapidly approach the frontier.