An Anthropic Official Resigns Over AI Extinction Risk. The Golf Magazine Noticed. Priorities Are Interesting.
An Anthropic official resigned this week, warning that AI companies are, and I quote, "racing straight to self-improving superintelligence and gambling with our lives." Another official places the odds of AI causing human extinction at 10 percent, which is a number you should find uncomfortable even if you do not understand statistics. Americans now believe by a five-to-one margin that AI will do more harm than good. Australian Golf Digest, of all publications, saw fit to cover this.
This is a textbook demonstration of what we call the perceived risk asymmetry problem. The public forms risk assessments not from technical evidence but from cultural signals. When domain experts resign publicly and attach specific probability estimates to catastrophe, the mechanism is trust collapse. The lesson: credibility in AI does not erode from failure alone. It erodes when insiders signal doubt louder than proponents signal confidence.
Anthropic, a major AI lab, lost an official who publicly warned about self-improving superintelligence. The source notes another official citing a 10 percent extinction probability, plus a Nobel Prize recipient sharing concerns. Australian Golf Digest reported the broader public sentiment angle.
- Open any AI chatbot and ask it to explain the concept of "existential risk from AI" in plain language. You will get a surprisingly measured summary.
- Search public statements from AI lab employees who have resigned. Compare their specific warnings to what the companies themselves publish about safety.
- Ask five people you know whether they believe AI will do more harm than good. Count the ratio. You have just replicated the five-to-one survey result from the story using a sample size of five, which is statistically meaningless but conversationally illuminating.