James Freeman spent the cable years training business-class readers to experience voluntary government cooperation as tyranny. His Wednesday Journal column on Trump’s AI executive order runs that training exercise again — only now the rhetoric has to bend around the awkward fact that Donald Trump, not Joe Biden, signed the order. What results is a text perfectly content to contradict itself if the contradictions buy donor-class permission a few more weeks. Freeman manufactures a regulatory phantom to justify handing the AI economy to unaccountable corporate monopolies. We used to load the vocabulary this way for the donor class; the column walks through the technique deployments as they appear.
This column recently lauded President Donald Trump’s deregulatory zeal and warned that his extremely wise decision to reject Biden policy and endorse the freedom of Americans to develop artificial-intelligence technologies was in danger of being reversed. Sadly the president is now suddenly looking less zealous.
The opening move deploys the hagiographer-in-miniature register — the “deregulatory zeal” variant, WSJ §A.1. Freeman opens by affirming Trump’s “extremely wise decision” to reject the Biden-era AI safety framework, which sets up the column’s preferred universe: one in which the federal government simply stays out of the way of capital. “Sadly the president is now suddenly looking less zealous” — the word “sadly” does the permission-structure work. It signals to the Journal’s business reader that Freeman is still on their side while the target has shifted from “Biden’s overreach” to “Trump’s disappointing timidity.” The apparatus stays aimed at Washington; the target-of-the-week changes. In the cable years, we called this “loading the vocabulary.” The operator’s goal is never to debate policy substance; the goal is to pre-load the cognitive frame so that any government interaction reads as coercion.
That was the right instinct. Unfortunately the new compromise poses a risk to U.S. technological vitality and prosperity.
Frame-engineered relabeling — WSJ §A.1 — operates here through “the new compromise.” Trump’s executive order asks tech companies to voluntarily give the government up to thirty days of advance access to new frontier AI models before public release, with no mandatory licensing, pre-clearance, or permitting requirement. The order’s own text is bluntly explicit: “Nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement.” Freeman cannot call the order “regulatory overreach” because Trump signed it and because the order’s text explicitly disclaims any licensing authority. “Compromise” fills the gap: it lets the reader feel the betrayal without requiring Freeman to produce a regulation that isn’t there. The column is not defending an actual freedom. It is defending a protection racket for developers who want to deploy software without anyone asking what it does first.
AI is now vulnerable to Washington regulators who have a long, sad history of imposing costly mandates that were never enacted in law, never explicitly approved by Congress.
The threat-inflation closer arrives early. “Vulnerable to Washington regulators” with “a long, sad history” is the WSJ page’s oldest trope: the distant, illegitimate, unaccountable bureaucracy strangling innovation. “Costly mandates that were never enacted in law” is doing the heavy rhetorical lift, since the order contains no mandates and no costs. The reality of the executive order in question — a voluntary request for a 30-day preview of models — gets erased by the language of “costly mandates.” The reader who does not click through to read the order hears only “mandates” and “Washington regulators” and completes the inference the column has supplied.
What are the companies supposed to do when government officials respond to each new model with a list of bureaucratic suggestions allegedly intended to improve the software?
The faux-naïf operator’s question — a technique so routine in this column’s register that it barely needs a catalogue entry. Freeman knows exactly what the companies are supposed to do: they are supposed to take the thirty-day voluntary review, nod politely at the government’s suggestions, and release the models unchanged, because the order contains no enforcement mechanism. The question is not a question; it is the permission structure that lets the Journal’s business reader experience a non-binding government request as a hostage situation. Freeman’s column has been building this audience-management move for years: whenever Washington asks industry for anything, frame the ask as the thin end of a regulatory wedge, regardless of what the text says.
Obviously there are those who fear this technology, but even the fearful seem to understand its potential benefits.
The dismissive-gesture strawman — the NR variant, but fully at home on the Journal’s op-ed page. “Those who fear this technology” is a category that includes, as of this week, a majority of respondents in the Economist/YouGov poll Freeman himself will cite in the next three paragraphs. But the strawman lets Freeman bracket those fears as irrational — “even the fearful” — before presenting the poll results. The rhetorical work is done before the data arrives.
According to the column, a new Economist/YouGov poll finds survey respondents in a sour mood about the country and much else. When it comes to AI in particular, a plurality see a negative impact on the economy.
Yet the poll also asked participants to respond to the following statement: “Artificial Intelligence will bring positive changes to the way we work and live.” A full 57% of registered voters in the survey agree, while only 30% disagree.
Here is the dual-message architecture at its most naked. To the populist reader: “a plurality see a negative impact on the economy” — validation of their anxiety. To the technocratic reader: “57% agree” that AI will bring positive changes — the credentialed numbers that make the column citable. The two sentences sit adjacent and contradict each other. Freeman does not resolve the tension; he does not need to. The populist reader stops reading at “negative impact”; the technocratic reader cites “57% agree.” Both audiences get what they came for, and the column incurs no cost for the contradiction. The operator’s-eye view on polling is straightforward: find the slice of the data that aligns with the donor class’s preferred policy — unregulated deployment — isolate it from the broader economic anxiety, and present the slice as the consensus. The 57% aren’t contradicting the sour mood; they are experiencing the sour mood precisely because the systemic risks of AI — job displacement, wealth concentration — are exactly what the pollsters captured in the first question. The piece is cherry-picking public optimism to excuse corporate recklessness; it is a shell game that distracts the reader with the promise of personal benefits so they don’t notice they’re footing the systemic bills.
But these positive changes will not happen if AI is directed by the federal government.
The pivot sentence from the poll to the editorial’s standing position. The technique is the unsupported-assertion-as-inevitability — the “of course” / “obviously” marker’s less-visible cousin. “These positive changes will not happen” is asserted as causal necessity, not argued. The premise being smuggled — that voluntary government review of AI models constitutes federal “direction” of AI — is precisely the claim the column has spent the first half trying to establish by relabel rather than by evidence. The sentence reads as conclusion; it functions as premise.
Alex Tabarrok writes at Marginal Revolution with a reminder of the importance of America’s services economy and the possibilities for an AI export boom.
Freeman then pivots to the technocratic-credential ledger, plus the broader “study shows” pattern. Tabarrok is cited by name, his institutional home at Marginal Revolution implied as authoritative, and the economist Richard Baldwin is quoted through him. The two experts together establish the column’s preferred economic frame — AI as an export-growth engine — as the serious-economist consensus, while the fact that neither Tabarrok nor Baldwin is quoted on the executive-order question goes unmentioned. The reader absorbs the authority and maps it onto the column’s thesis: deregulation works; “compromise” threatens export growth. The linkage is associative, not logical.
Are We Negotiating with an Ayatollah or an AI-atollah?
The section break is the column’s formal acknowledgment that it has no coherent structure. One subhead covers the AI executive order; the next subhead — a pun on “Ayatollah” and “AI-atollah” — opens a new subject entirely: whether Iran’s new leader is alive. The transition does not attempt to link the two subjects. The abrupt pivot to Iran is not a structural flaw; it is the holding company rotating to service its next demographic tenant. One paragraph you are reading about AI export projections; the next paragraph you are reading about Ayatollah Mojtaba Khamenei’s public invisibility. The BOTW column format — Freeman’s weekday “Best of the Web” franchise — organizes itself as a series of short, disconnected items, and the discontinuity between the AI item and the Iran item makes visible what the format usually conceals: there is no argument connecting the pieces. The column is a scroll of permission slips, not an essay. The AI piece ends when Freeman runs out of donor-class signaling; the Iran piece starts because the format requires another item. The rhetorical work of the “AI-atollah” pivot — a form of manufactured urgency, Playbook §5.14 — connects a voluntary domestic model-access request to high-stakes geopolitical maneuvering with Iran. This is the standard apparatus move: corporate lobbying dressed in the language of geopolitical survival. The White House order text contains zero references to foreign adversaries or Middle Eastern negotiations; the linkage is an editorial fabrication designed to dress corporate lobbying as statecraft.
Catholic archbishop of New York Ronald Hicks writes in First Things on the rise of physician-assisted suicide.
The final item drops a name-check of Archbishop Hicks; the elevation of a friendly institutional voice without direct argument is the signal — no quotation is needed because the name alone confirms the Journal page’s alignment with Catholic moral authority on end-of-life questions. It sits disconnected from AI, from Iran, from anything that preceded it except the column’s standing audience-management alignment: the Journal’s business-conservative readers are also the audience for this signal. No argument is offered because none is needed; the name-check is itself the permission that the Journal’s page stands with the archbishop, and the reader who stands with the archbishop is reading the right page.
Freeman’s column runs four items — AI panic, AI polling, Iran palace intrigue, Catholic bioethics — held together by nothing except the reader’s pre-existing alignment with the Journal’s editorial commitments. Each item supplies a permission structure to a different audience segment. The AI-executive-order item tells the donor class that even Trump cannot be trusted to hold the line against Washington; the polling item tells the populist that their AI anxiety is respected while telling the technocrat to cite the 57% number at their next meeting; the Iran item tells the national-security reader that the Journal is monitoring the Khamenei succession; the assisted-suicide item tells the religious reader that the page stands with the archbishop. None of the items depend on each other, and none of them supply evidence for the others. The column is not an argument. The column is a holding company for four identity-reinforcement operations, and the format — Best of the Web as a daily scroll of linked-but-disconnected items — is the corporate structure that makes the holding company look like a column.
So here is what the column actually does, taken together. It takes a voluntary, 30-day pre-release review — a transparency measure so mild the word “request” appears in the executive order — and inflates it into a bureaucratic takeover of the American economy. It polls the public, isolates the one metric that justifies the status quo, and discards the rest. It ties a domestic lobbying victory to foreign adversaries to make corporate deregulation sound like a patriotic duty. The piece doesn’t want a debate on AI safety. The piece wants a permission structure. It wants the reader to feel that any oversight at all is a betrayal of freedom.
What it declines to say — what the format’s very structure prevents it from saying — is that the order Freeman is criticizing as a sellout was the product of weeks of negotiation between the administration and the very tech CEOs whose interests Freeman’s column purports to represent. The companies asked for the order. Freeman’s readers are not told this. The gap between what the column knows and what the column says is the gap the Journal’s op-ed page has been monetizing for decades. Freeman just runs the register.
The reality is much simpler: the column is asking the public to voluntarily surrender their oversight of the most powerful technology in human history because the people building it want the public to trust them without ever having to explain what they’re building — the exact asymmetry the column’s regulatory phantom is designed to protect. The whole thing amounts to a simple donor-class racket.
— Phukher Tarlson