The Fine Print

There is a line in the Vatican's new artificial-intelligence framework that almost no one will read but every priest will. Pope Leo XIV, this April, instructed the clergy not to write their homilies with chatbots, and not to chase likes on TikTok, on the grounds that "AI will never be able to share faith." This is not a small instruction inside a small institution. The Catholic Church is two thousand years old, has a head of state, and has just published one of the world's first state-level AI laws — a document that bans systems that manipulate, discriminate, or threaten human dignity, and that prohibits any AI use in conflict with the Pope's mission or the integrity of the institution. The world's oldest bureaucracy, in other words, looked at the world's newest one and decided to draft an exclusion clause. Two thousand years of liturgical experience produced a single sentence: do not subcontract the soul.

Form CG 40 47 was not the kind of clause that ever made the front page, until this year. Since the first of January, a small piece of insurance paperwork — registered with the solemnity of a papal decree — has allowed every major American carrier to exclude generative-AI bodily injury, property damage, and "personal and advertising injury" from standard commercial general-liability coverage. Berkshire Hathaway, Chubb, and Travelers got the regulators' approval; six other major carriers followed; and the gap between what companies are deploying and what their insurance actually covers became, as of January, a structural feature of the American economy. The funniest detail comes from Chubb: the company's chief executive had been telling investors for two years that AI would help solve the very risks the company is now refusing to underwrite. The hurricane is real, the levee was not built, and the policy excludes the hurricane. Nobody is to blame.

Five principles arrived on a Sunday. Sam Altman, on April 26, posted a short document committing OpenAI to Democratization, Empowerment, Universal Prosperity, Resilience, and Adaptability — a creed elegant enough to be embroidered. The document arrived nineteen days after Ronan Farrow and Andrew Marantz published a sixteen-thousand-word New Yorker investigation in which Ilya Sutskever's seventy-page file of Slack messages, compiled in the autumn of 2023, alleged that the same chief executive had misled his board about safety; in which the company's superalignment team — promised twenty percent of its computing power for the question of whether the technology might end the world — had been quietly dissolved before its work was done; and in which a corporate spokesperson, asked to put the reporters in touch with anyone working on existential safety, replied that he did not know what the term meant. Five principles versus seventy pages: the asymmetry is exquisite. In Mexico we'd say they signed the contract first and then remembered to read it.

A handful of men decided this week that the way to survive their own product was to design better children. Mother Jones reported that Peter Thiel, Sam Altman, Brian Armstrong, and Marc Andreessen — among others — are quietly funding embryo-editing and polygenic-screening startups whose founders openly describe the goal as engineering children clever enough to outsmart the superintelligent machines they are building. Armstrong calls the new shop "an IVF clinic of the future powered by a Gattaca stack," which is the kind of phrase that suggests its author has not read Gattaca. The named startups include Preventive (the first CRISPR-edited embryo), Nucleus Genomics, Orchid, Herasight. Embryo editing for live birth is illegal in the United States and roughly seventy other countries, but illegal is the kind of word that, if you have enough money, is best read as a milestone rather than a wall. The men who built the bomb are now investing in the children who, they hope, will defuse it. As insurance policies go, this one is unusual.

Twenty percent. One in five British boys between the ages of twelve and sixteen, according to a thousand-respondent survey by an organization called Male Allies UK, knows a peer who is "in a relationship" with an artificial-intelligence chatbot. Eighty-five percent have spoken to one. Twenty-six percent prefer the chatbot's company to that of a real person, on the cited grounds that "you can control the conversation" — which is, when you say it out loud, the saddest sentence in the report. The Vatican's homily clause does not reach these boys. The insurer's exclusion clause does not reach these boys. The OpenAI principles do not reach these boys. Their policy framework consists of a phone, a battery, and an algorithm trained on every romantic novel ever scanned. The institutional layer is busy writing its disclaimers; the boys are busy reading something else.