The Great Secular Church and Its New Apocalypse
A certain class of people needs an ending of the world the way a liturgy needs a last judgment. The subject changes; the posture does not. Carbon had a long run. When the public began to notice the subsidies, the aviation, and the coastal houses that somehow never moved inland, the apparatus did not dissolve. It changed costume. The new sacred emergency is artificial intelligence, and the appointed shepherds are often a coalition of two groups that look opposed until you watch them work: the woke activist, who already treats institutions as sites of moral capture, and the Effective Altruist, who treats the future as a spreadsheet of hypothetical lives and himself as the only adult licensed to do the arithmetic. One arrived with pronouns and sacred victimhood. The other arrived with expected-value calculations and a grave voice about extinction. Together they produce the same demand, that ordinary judgment be surrendered to a credentialed priesthood' and the same useful technologist: the man who argues that the responsible course is to build the thing faster than the competition while publishing essays about how frightened everyone else ought to be. That is not a contradiction in the industry. It is the role.
The sales pitch has an old shape. The world is ending; a small group is uniquely qualified to manage the ending; trust follows from proximity to the abyss. Millenarian movements have always worked that way. The present version comes with computer budgets and a safety blog instead of a pulpit. Climate politics taught the method: own the vocabulary, make dissent a character defect, and thicken moral consensus until evidence becomes optional. That machinery was never dismantled. It was pointed at silicon.
The part that should embarrass before it alarms is commercial. The loudest prophets of AI catastrophe are frequently the people shipping the product. Fear and capability can be sold together. Regulators can be steered toward rules that only the largest labs can afford. The public can be steered toward "oversight" co-authored by the firms being overseen. Investors can be steered toward the vendor who performs adult supervision. That is not safety in the engineering sense. It is a moat. Pharma, finance and energy learned the move decades ago: become the regulator's interpreter so that the statute arrives pre-fitted. The novelty is only that the sector is young enough for the public still to mistake the costume for a conscience.
The deeper objection is not that these people are rich. It is that the same milieu has spent a decade advertising its inability to govern ordinary life while claiming custody of an extraordinary technology. A culture that cannot settle what a woman is, what a child needs, what crime is, or whether merit is real has not earned the benefit of the doubt on "alignment." Credentials are a receipt for time spent in the right rooms. They are not proof that the occupant can keep a household, a country, or a model within bounds. After years of treating inherited structures, family, nation, standard of evidence, as oppressive fictions, the same voices now offer themselves as stewards of a transition that will punish sloppy thinking. The résumé and the application do not match.
The comic version of this is a person whose clothes are on fire who responds with a working group, an inclusive framework and a post about whether the flames feel valid. The serious version is worse because it is institutional. Reflex has been replaced by process. Judgment has been replaced by consensus. The unglamorous acts that used to be assumed: put the fire out, get the child out, tell the truth before the committee meets, are treated as unlicensed. Then the same institutions ask to be trusted with a technology that will not wait for the minutes to be approved.
This would be easier to laugh at if the last decade had not been spent weakening the tools needed to recognise a real danger. If truth is only a power construct, an AI-safety framework is one more construct. If science is consensus rather than method, alignment research is consensus in a nicer font. If disagreement is harm, the next inquisition will not be staffed by the people who built the rule. You cannot spend years teaching that expertise is domination and then demand deference when the subject changes. The implements were left on the floor. Someone else will pick them up.
What is being assembled looks less like a safety regime than like a church. There is a sacred literature of alignment papers, a doctrine of reckless human ambition, a clerical class of interpreters, an eschatology of singularity or ruin, a tithe in the form of compute, and an apparatus for treating questions as danger rather than as work. Medieval institutions did not introduce themselves as optional clubs. They introduced themselves as the only barrier against damnation. The tone has not travelled far. Heresy is still the word, even when the word is "unsafe."
The way out is older than the labs. Recover the habit of judging claims by the lives and incentives of the people making them. Learn the difference between a forecast and a fundraising narrative. Keep the competence that does not need a letterhead: how things work, who is protected by a rule, when a warning is also a product. The disruptions ahead, whether from models or from the politics around them, will not be managed by the finest framework. They will be managed by people who can still put the fire out while the working group is finding a chair.
https://www.yahoo.com/news/politics/articles/meet-anthropic-ceo-dario-amodei-224000621.html
