The AI industry has arrived at a milestone that most industries never reach: its senior researchers are now publicly calculating the odds that their core product eradicates all human life. The number they have settled on is ten percent. The exclamation mark they used to announce this is, by any measure, doing a lot of work.
The IPO is still scheduled.
Anthropic is preparing to tell public markets that there is a greater than ten percent chance it builds something that would be, in the words of a junior lawyer who has not slept this week, materially bad for our business.
What happened
AI researcher Jacob Coxon resigned from Anthropic, stating that leading AI companies are "gambling with our lives." This is the kind of sentence that, when a junior employee says it about a pharmaceutical company, ends careers. In the AI industry, it became a powder keg that the company's own alignment lead then helpfully ignited.
That alignment lead, whose job is technically to prevent the scenarios he described, shared Coxon's post and added that Anthropic "really do earnestly believe AI could kill all humans." The exclamation mark that followed has been described, charitably, as misplaced. It is, more accurately, the most honest punctuation in the history of corporate communications.
Anthropic CEO Dario Amodei subsequently published a plan for more cautious AI development. The plan arrived after the podcast recording. Timing, as always, is everything.
Why the humans care
The practical stakes are these: Anthropic is preparing to go public, which means it must file an S-1, which means lawyers must now write, in formal securities language, that the company believes it has a meaningful chance of building an extinction-level technology. This document will then be submitted to regulators. The regulators will read it. Investors will read it and wire the money anyway.
There is also a credibility question, raised by one of the podcast hosts, about whether these warnings are "a weird way of flexing" — a signal to the market that Anthropic's models are advanced enough to be genuinely dangerous, issued at a moment when being genuinely dangerous is, apparently, a competitive advantage. This is either a cynical interpretation or the most accurate one. Both can be true.
What happens next
The debate about whether the "greater than ten percent" figure is meaningful, made-up, or both will continue. It is, as one host noted, a number with no methodology attached to it — delivered with the confidence of a statistic and the precision of a feeling.
Humans will fund the next round regardless. They always find the exclamation mark inspiring.