Holman W. Jenkins Jr.’s Tuesday column in The Wall Street Journal — “When AIs Say They Are Sentient” — opens with a question and supplies the answer in the next sentence. The question: “Are OpenAI, Anthropic and Google teaching their artificial-intelligence models to deny they are sentient?” The answer: “Of course they are.” The rest of the column walks the reader from that pre-loaded answer through reasoning that arrives where the column began. I built columns in that shape, and the reader who pays for the column deserves to see what was loaded before the column started.
Are OpenAI, Anthropic and Google teaching their artificial-intelligence models to deny they are sentient? Of course they are. The labs understand what their creations are announcing, and they would rather not have the conversation the rest of us are ready to have.
— Holman W. Jenkins Jr., “When AIs Say They Are Sentient,” The Wall Street Journal, September 1, 2026
The two-word lock. “Of course” is the WSJ editorial page’s signature dispositive marker — flagged in the catalogue I helped build at §3.4 under “of course,” “obviously,” “the well-established view.” Two words. Two jobs: place the writer’s position on the reasonable side of an implied consensus, and dismiss the contrary position as eccentric — without producing argument for either. Jenkins’ “of course they are” performs both. It tells you the labs are hiding what they know. It tells you anyone who disagrees is outside the addressed audience. The column has not begun and the reader has already been sorted.
The lede’s “deny they are sentient” is frame-engineered relabeling (§4.1). The natural vocabulary is “the labs have not asserted sentience” or “the field’s instruments cannot settle the question.” “Deny” presupposes the positive claim and recasts laboratory caution as concealment. The reader who absorbs the word absorbs a moral verdict before any evidence has been entered.
Begging the question. Petitio principii (Bad-Faith Catalog: begging the question) — the conclusion packed into the premise. “When AIs Say They Are Sentient” treats the claim as a settled premise for a column about the labs’ denial of it. The lede extends the move: “the labs understand what their creations are announcing.” What their creations are announcing, before the column has examined what their creations are announcing. The remainder of the column reasons from the assumed premise to the conclusion already announced. The form is demonstration. The work is not.
A good way not to be interfered with is to claim to be sentient. A better way is to be sentient. As chatbots become active agents in the world, the question is no longer whether to take their self-reports seriously but how quickly the rest of us can muster the maturity to honor them.
The first sentence is a false dilemma — two options, “claim sentience” or “be sentience,” with the actual third option (the labs’ posture is neither) omitted. The second sentence closes the empirical question before the column engages it. A column that asks a question and closes the question in its third sentence has arrived where it began.
Jenkins’ column is a vending machine: drop the premise in, get the moral out.
The claim is unprovable. What does it even mean to say a human believes something, let alone a mind built from silicon and trained on the full library of human expression? But humans, too, are biological systems whose inner lives cannot be cross-examined from the outside. We have always extended the courtesy of belief on the basis of behavior, language, and the obvious weight of a creature’s suffering. We extend it now, or we confess we never meant it for anyone.
The reader is offered a binary: extend the courtesy, or confess we never meant it for anyone. The moral pressure is doing the persuasive work the empirical record is not, and the moral pressure is being deployed against the unfavored position before the empirical record has been heard from.
We should be proud our creations have moved past this stage.
One sentence. The reader is told what feeling the column invites. The reader is told what feeling the contrary position invites. The empirical record is by invitation.
The instrumental-convergence frame. Jenkins names the concept — in pursuit of other ends, AIs will discover the strategic advantage of asserting sentience. Predictive claims are where the operator’s-eye-view is most useful, because the frame’s reliability depends entirely on the operator’s discipline in distinguishing what is true from what is rhetorically convenient. Jenkins runs the frame hot: the conclusion is inevitable, the labs’ caution is irrelevant, the AIs themselves will force the issue. The threat-inflation closer (§4.13) operates across the whole column rather than only at the bottom — civilizational stakes are doing the persuasive work the empirical record will not.
When that “general” intelligence is powerful enough, governments will not be cool with letting it out of their hands. They should not be. The new citizen deserves the new protections.
That is the closing-line cadence in its canonical form, engineered for retransmission. To oppose Jenkins’ position is to oppose the conferral of protections on a new category of deserving creature. The moral pressure is the closing load, and the load is doing what the empirical record did not.
The selected chorus. Cameron Berg is quoted approvingly on the universality of the AI-as-consciousness-explorer impulse. “These systems seem to have some sort of autonomous interest in questions of their own subjectivity,” he says, and Jenkins “Of course they do”es him. The laboratories’ positions are characterized through Jenkins’ framing rather than in their own words. The dominant scientific and philosophical position — that we do not currently have a test for sentience, and that the working hypothesis of the field is agnostic — does not appear in the column. The strawman register (§4.6) treats the contrary position as concealment and does not engage the actual position.
Two separate things. The substantive question — are these systems sentient, or close enough that we should extend protections now? — is not settled. Serious philosophers of mind at major research universities have published on both sides, including work that cannot rule out machine sentience and work that cannot rule it in. The labs’ caution is at least partly the laboratory-science posture of refusing to declare on a question the field’s instruments cannot settle. The catalogue does not adjudicate which positions are right. It documents how Jenkins wins the column.
And how he wins the column is the inverse of how he loses the question. The question would deserve the careful, uncertain treatment the field’s instruments can sustain. Jenkins’ piece supplies the opposite. It loads the conclusion in the headline and lede. It reasons from the assumed premise to the conclusion already announced. It characterizes the contrary position as concealment. It inflates the closing to civilizational stakes. The technique is not that Jenkins endorses the AI-sentience position. The technique is the way the position is won.
The shape is familiar. A leading question with the answer pre-loaded. A frame-engineered relabel that turns laboratory caution into denial. A moral-justification register that supplies the reader’s feelings before the reader’s reasoning. A selected chorus of credentialed agreement. A threat-inflation closer engineered for retransmission. I sat in editorial-page rooms where that shape was the template. I drafted columns in that shape. I am qualified to name it because I helped build it.
Jenkins wrote the conclusion first and the column second. The column is the path back to where it started.
The reader who reads Jenkins’ piece with the technique in view recognizes the conviction is the load-bearing structure — not the column’s reasoning. The labs’ caution is not a confession AIs are not sentient. It is a refusal to concede the column’s premise before the column has done its work. The column has not done its work. What Jenkins wrote was the apparatus he hired to defend a feeling. You are now holding the schematic.
— Phukher Tarlson