There is something touching about the AI engineer who quits a frontier lab because he believes the machines are going to kill us all within ten years. One imagines him striding out of the office with the solemnity of a monk leaving Rome, not because the church is too corrupt, but because the comet has been sighted and nobody in management has updated the sprint board accordingly.
Jacob Coxon’s resignation from Anthropic, as reported by outlets including The Guardian, New York Times, CBS News, and ABC, is presented as a warning from inside the temple: the people building advanced AI, he says, privately believe it could destroy humanity by the end of the decade. The claim has a certain cinematic grandeur. It also has the familiar structure of millennarian prophecy: the end is near, the faithful insiders know it, the institutions are blind, and repentance must begin before the appointed hour strikes.
The Secular Rapture
The apocalyptic preacher traditionally stands on a street corner with a sandwich board reading, “The End Is Nigh.” The AI-risk engineer has upgraded the format: he posts a thread, invokes recursive self-improvement, and supplies a confidence interval. Yet the emotional architecture is much the same. There is a fallen world, a forbidden power, a priesthood that understands the danger, and a deadline just close enough to terrify but just distant enough to escape falsification until the next forecast cycle.
Like the religious fanatic predicting Judgment Day, the AI doomer does not merely warn of risk; he speaks as if history has already been indicted. The companies are not making questionable strategic choices; they are “gambling with our lives.” The engineers are not debating uncertain scenarios; they are custodians of an awful truth. The rest of us, benighted villagers that we are, continue buying laptops, answering emails, and asking chatbots to rewrite memos, unaware that we are apparently living in the final act of a techno-theological disaster film.
The Miracle Without a Mechanism
The most comic feature of this prophecy is not its severity but its vagueness. The machine will kill us all. How? Well, it will become superintelligent. And then? It will be able to hack things. Which things? Perhaps all things. And after that? It will acquire resources. Which resources, through which institutions, under what constraints, with what physical embodiment, and against what countermeasures? At this point the sermon turns misty. The dragon is unmistakably coming; details of the dragon’s supply chain remain under peer review.
This is where the resemblance to end-times religion becomes irresistible. The believer knows the apocalypse is imminent, but the causal pathway is oddly elastic. Fire from heaven, plague, beast, singularity, hacked infrastructure, autonomous persuasion engine—choose your eschatological interface. The prophecy survives because it is not pinned down by a concrete sequence of events. If one path looks implausible, another emerges. The end is certain; only the mechanism is TBD.
The Ten-Year Rolling Apocalypse
The ten-year horizon is especially convenient. It is close enough to demand urgency, far enough to avoid immediate embarrassment, and flexible enough to be renewed as needed. Religious movements have long discovered the utility of this arrangement. The apocalypse must be close, because otherwise people will not repent; but it must not be tomorrow, because tomorrow has an irritating habit of arriving on schedule.
To be fair, none of this proves advanced AI is harmless. Powerful systems can create real risks, and serious safety work is plainly worthwhile. But a civilization should be able to distinguish between a risk argument and a revelation. “Something very bad could happen through some sequence of technical, institutional, and geopolitical failures” is a proposition to analyze. “We are racing toward machine Armageddon and only the initiated understand” is a mood, a liturgy, and possibly a recruitment strategy.
Against Apocalypse as Argument
The useful question is not whether someone can imagine a computer powerful enough to end the world. Human beings can imagine almost anything, including gods with thunderbolts, demons with contracts, and software with a grudge. The useful question is whether the prophet can describe a plausible chain from present systems to global extinction—step by step, institution by institution, failure by failure—without replacing explanation with awe.
Until then, the AI apocalypse sermon deserves the same skeptical courtesy we extend to every end-times proclamation: listen politely, ask for the mechanism, and notice when the answer arrives wrapped in incense, probability estimates, and a suspiciously adjustable deadline.
The post The Prophet of the Machine Apocalypse appeared first on Internet Governance Project.
