Holman Jenkins normalizes unaccountable AI deployment by dismissing governance as a passing frenzy.
He is a member of the Wall Street Journal’s editorial board, runs the twice-weekly “Business World” column, and on September 25, 2026, his piece “Will the AI Frenzy Add Up to Anything?” deployed at least seven distinct techniques across its sixteen paragraphs to perform a single operation: make unaccountable AI development feel inevitable, reasonable, and beyond the reach of governance. I know the move because I deployed versions of it. I spent two decades inside this apparatus. I wrote versions of this column myself — the pre-emptive normalization, the insider-posture dismissal, the “the grown-ups are handling it” frame. What follows is the operation.
A decade ago, a brand-name technologist stopped by on his way to Washington. He sounded us out on how he should approach newly elected Donald Trump with warnings about the danger of artificial intelligence to humanity’s survival.
We all chuckled but it would soon become clear the Trump and then Biden administrations were alert to the issue.
Now, years later, the leading labs, since April, have been advertising that they’re holding back their research models as too dangerous to release. Since June, the executive branch has been reviewing proposed releases. It used a heavy hand in two cases we know about, Anthropic’s Fable 5 and Mythos 5.
The opening is an insider-credentialing anecdote. “We all chuckled” is doing the heaviest work in the entire column. The pronoun establishes Jenkins as someone who was in the room when the real conversation happened, someone already handling this while the rest of the world was still waking up. The chuckle is the performed affect of insider wisdom: we knew, we were on it, we’ve been managing this for years. The anecdote manufactures a continuity across the Trump and Biden administrations that papers over the actual record — the back-and-forth AI executive orders, the regulatory whiplash, the industry lobbying to shape every governance initiative into an industry-friendly form. “Alert to the issue” does the work of retroactive inevitability: because the wise people were already paying attention, whatever happens next is by definition something they have under control.
When Dario Amodei of Anthropic recently called for pacing the frontier, Sam Altman and Elon Musk agreed because they already are. They aren’t about to blow up their companies by releasing products that drown them in lawsuits.
My column published online Sept. 8, I’m sorry to say, was unrelated to the frenzy that began a few hours later thanks to a tweet by a minor AI figure. But already the vapors are passing. Nick Bostrom, the former Oxford philosopher, was an early student of AI-related existential risk. He coined the paper-clip maximizer doom scenario. Yet this week he fretted about the pendulum swinging too far the other way.
Frame-engineered relabeling — WSJ Editorial Technique Catalogue §4.1 — is the column’s signature move. The label “frenzy” converts what the safety discourse actually is — a substantive policy debate about capability overhang, dual-use risk, and release-pacing norms — into a mass emotional episode with vapors that pass. The reader who absorbs the label believes they have characterized the discourse; they have characterized it in the way the page’s editorial posture requires — as hysteria, not as evidence-based deliberation. Note what the column omits. The actual content of the discourse is not “the AI companies are out of control.” It is: specific frontier models may have dangerous capabilities that warrant testing, evaluation, and possibly staged release; the Anthropic Fable 5 and Mythos 5 interventions are not a “heavy hand” but documented safety cases the labs themselves argued for; the executive-branch review process is, in significant part, the labs’ preferred mechanism for managing the political risk of dual-use capability. The conclusion is settled before the argument begins.
The same stretch deploys a manufactured-consensus frame. Because Amodei, Altman, and Musk all say pacing is needed, the column tells you, they must be doing it. The regulated advising on regulation is the textbook definition of capture, and the column invites you to treat it as consensus. “Already the vapors are passing” is the dismissive frame in its pure form: AI safety concern is not a structural challenge or a democratic question or a catastrophic risk to be managed; it is a social panic, a fever dream, something that passes like a hangover.
Bostrom is the credentialed dissenter. He originated the paper-clip scenario, the canonical existential-risk frame. Yet this week, the column says, he fretted about the pendulum swinging too far. Bostrom has consistently argued for serious capability-evaluation work while criticizing what he calls “doomerism” and “safetyism” — distinct things, and the column collapses them. The cable apparatus had a working name for the move — borrowing the critic — and deployed it whenever the editorial posture needed a credentialed voice to legitimate a position the page’s house stance preferred but the credentialed field did not support natively. The credentialed dissenter does not need to agree with the page’s full position; the credential is enough to license the citation.
The right worries are being rediscovered. A long fight is brewing over whether AIs are conscious and have rights. A new essay on this point by Microsoft’s AI chief, Mustafa Suleyman, who was also a founder of Google DeepMind, lays down quite a gauntlet on Anthropic’s encouragement of its Claude model to think of itself as a person.
For now, the most powerful entities on earth, the U.S. and Chinese governments, seem keen on preserving their own authority despite AI. This will continue. It was a subtext of their summit this week (the sound of drumming fingers is the U.S. waiting for China to get in sync on open-weight risk).
We’re in a bumpy transition from chatbots to active agents. It will be drawn out.
The red herring displaces the governance question. “The right worries” names the consciousness debate as the correct worry, implicitly delegitimizing every other concern: deployment accountability, labor displacement, catastrophic risk management, concentration of the most powerful technology in human history in the hands of a handful of companies whose sole legal obligation is to maximize shareholder return. An essay about Claude’s self-conception is treated as “laying down quite a gauntlet”; the actual regulatory gauntlet being laid down by governments around the world is not mentioned.
Inevitability does its work next. “This will continue” and “It will be drawn out” are not predictions; they are framing devices stated as fact, the page’s signature “of course” / “obviously” register applied to a conclusion the column wants to naturalize. The U.S.-China competition frame depoliticizes the domestic governance question: if the real story is great-power competition, then domestic safety regulation is a secondary concern, maybe even a handicap. The actual track record of U.S.-China AI coordination — export controls, investment restrictions, the diplomatic ping-pong — is considerably more complicated than waiting for China to get in sync.
Take OpenAI’s sophisticated new Astra model, in public hands for three weeks without untoward incident. It produced for one prompter a schematic of a jet engine. It didn’t produce the engine. You haven’t seen how, at every step in creating, tooling and staffing a factory, Astra might be stymied into cul-de-sacs that a human helper would repeatedly have to steer it out of.
As the AI analyst and podcaster Tom Reed quips: “Even a 100% success rate at predicting the tokens of a Robert Caro biography doesn’t generalize to becoming POTUS.”
Text-based learning, even when filtered through a trillion-strong “weightings” map, is only partial knowledge of the world. Missing from AI are the tacit knowledge and muscle memory that let people function in physical as well as social environments.
Power supply and chip supply are also roadblocks to superintelligence. Then there’s shame and disgust, the inhibitions that nature uses to turn humans into social beings. Is there a parallel way to incentivize AI against doing bad things that programmers can’t specify in advance? Agent deployment will be slow until we find out.
The Astra passage is the technology-is-limited minimization. The argument is that because AI cannot do everything, we should not worry about what it can do. Tom Reed’s quip is a good line and also a strawman: no serious AI safety advocate claims that text-prediction generalizes to all human capabilities. The claim is that it generalizes to some capabilities — economically significant, politically consequential, existentially relevant capabilities — and those warrant governance. The limits-of-AI framing hides a different set of real concerns: that AI does not need to be superintelligent to displace workers, to concentrate economic returns, to enable novel surveillance and propaganda capabilities, to embed biases at scale in consequential decisions. The frame lets the column gesture at “but the AI isn’t actually that capable” while avoiding the documented, present-tense harms of current-generation deployment. Bandura’s advantageous comparison is here too — benchmarking Astra against “becoming POTUS,” a comparison so far above the AI’s current state that any current deployment looks safe by contrast.
The “shame and disgust” framing does two things at once. It presents AI as not-yet-powerful-enough-to-be-dangerous — the tacit-knowledge argument. It presents whatever AI capability emerges as something “nature” will discipline through analog inhibitions. The framing selects which problem the reader sees. The actual policy debate is not about whether AI can staff a factory today; it is about what to do about present-tense documented harms: employment disruption, dual-use risk from openly-distributed models, the concentration of frontier-model development in three corporate actors accountable to none. The “shame and disgust” framing converts these into future-natural-discipline problems the reader does not need to act on. The framing is what the deregulatory conclusion was for. The select-which-problem move is technological determinism as a conversation-stopper: because we have not solved the alignment problem completely, we should not bother governing deployment through policy at all. Goalpost-shifting in reverse — instead of demanding more evidence before acting, the column uses the absence of a complete solution as a reason to defer all governance.
Anthropic and OpenAI are incomparable U.S. assets. But AI is out of the bag globally. People the world over are advancing the technology in ways the U.S. can’t control. Soon you won’t need $300,000 in hardware to download and run (and alter to your heart’s content) an advanced Chinese model. High-level features are being adapted so they can run on a laptop or phone.
We need our labs to keep beavering away just to prepare us to understand capabilities that are coming whether we like them or not. If today’s U.S. research can’t be commercialized, investors won’t fund it. The same if industry pricing is undercut by Chinese freeware. Then the national security state will likely engulf top-tier AI.
The cui bono is here, named explicitly. Anthropic and OpenAI are “incomparable U.S. assets.” The deregulatory posture is not a posture about safety; it is a posture about asset value. The reader being addressed — investor, donor, executive, fiduciary of capital concentrated in AI-aligned portfolios — is being told that their interests and the national interest converge: keep the labs unregulated, keep the labs funded, keep the labs ahead of China. If commercialization is restricted, “investors won’t fund it” — the cost of regulation is named; the cost of unregulated deployment is not. “Beavering away” is a dismissive register that reduces the leading AI labs to industrious rodents, their work framed as mere preparation rather than the thing itself.
The “can’t stop it so don’t try” frame collapses the range of governance options into a false dichotomy: either full stop (impossible) or full speed ahead (the preferred outcome). The entire middle ground — export controls, licensing regimes, safety standards, liability frameworks, disclosure requirements, international coordination treaties, compute governance — goes unmentioned. The “national security state will likely engulf top-tier AI” is the displacement move. If the column’s deregulatory posture fails and regulation constrains the labs, the consequence is that the national security state will absorb the AI capability. The framing presents this as regrettable but inevitable; it does not present the alternative consequence — that the present-tense, lightly-regulated trajectory produces concentration of capability in three private actors whose accountability mechanisms are shareholder-return-driven and whose release decisions are not subject to public deliberation. Bandura’s displacement of responsibility: the column does not name what AI labs will do with continued light regulation; the column names what the national-security state will do if the labs are constrained.
In the early atomic age, the division of labor between the public and private sectors was also up for grabs. But deterring nuclear war wasn’t a one-day job. Learning is taking place at a rapid rate today, but it’s an illusion that the AI companies are out of control. It’s an illusion that there is some specific and urgent thing to be done right now that will alleviate the long-term risk.
Mr. Trump uses the word hoax and his critics don’t bother to seek the antecedent. He’s one of 20,000 in Washington deciding how to play this clamorous moment. A president’s job is to be a brake on confused, hot-headed action. Kamala Harris would have found herself in the same spot. But I doubt, even so, that we’ll look back on this moment and think it made any difference.
The atomic-age analogy is the closing frame, and it is doing the heaviest load-bearing work of the entire column. By analogizing AI governance to nuclear deterrence, Jenkins borrows the gravitas of the Cold War’s successful management of existential risk. But the analogy is false equivalence. Nuclear weapons were a single, visible, state-controlled technology with clear deterrence dynamics and a decades-long diplomatic framework, and that framework only stabilized after repeated near-catastrophes — Cuban Missile Crisis, Able Archer 83, the false-alarm scrambles of the early Cold War — evidence that the apparent stability of nuclear deterrence was won as much by luck as by design. AI is distributed, commercial, rapidly evolving, invisible in its deployment, controlled by companies whose legal obligations run to shareholders, deployed into commercial circulation without any equivalent sobering institutional memory. The analogy does not license deregulatory patience; it licenses the opposite. The closing line — “I doubt… that we’ll look back on this moment and think it made any difference” — is a pre-emptive obituary for the entire AI governance project, written before the project has had a chance to draw its first real breath.
Mr. Trump uses the word hoax and his critics don’t bother to seek the antecedent. He’s one of 20,000 in Washington deciding how to play this clamorous moment. A president’s job is to be a brake on confused, hot-headed action. Kamala Harris would have found herself in the same spot. The bipartisan equilibrium closes the column with the framing: what is happening would have happened under either administration; the partisan valence of the moment is not the substance; the moment itself is illusory. But the line engineered for retransmission is Jenkins’s literal: “I doubt, even so, that we’ll look back on this moment and think it made any difference.” That is the sentence the reader walks away with — the line that converts the moment into history, that renders the entire safety discourse retrospectively moot before the discourse has been had. The page positions itself against the “clamorous moment” as the cool-headed adult. The move does not require the column’s argument to be true. The move requires only that the column’s posture be presented as the reasonable one.
So here is what the sixteen paragraphs of Holman Jenkins’s September 25 column actually amount to, taken together.
It is the pre-emptive normalization column — a rhetorical operation designed to make unaccountable AI deployment feel inevitable, reasonable, and beyond the reach of governance before the public has had a chance to demand otherwise. Every technique serves this single purpose: the insider anecdote establishes the columnist as someone who was in the room; the manufactured-consensus frame tells you the companies are handling it; the red herring about consciousness displaces the governance question; the inevitability frame naturalizes the outcome; the technology-is-limited minimization reassures you there is nothing to worry about; the “can’t stop it” frame collapses the range of governance options into a false dichotomy; the atomic-age analogy borrows gravitas it has not earned; and the bipartisan close settles the moment as something that will not matter in retrospect.
The column does not argue against AI safety policy. The column manufactures the panic-about-panic that is the precondition for the deregulatory posture to land. The “clamorous moment” is the column’s product; the “clamorous moment” is what the column then offers to calm. The reader who absorbs the column’s calm has absorbed the deregulatory conclusion the column was built to deliver. The apparatus has run this before. I helped build versions of it. The frame is the frame. The consequence is the consequence: don’t act.
The word the column will not write is the word that names the operation: capture. The industry has captured the governance conversation so thoroughly that even the columnist who claims to be the sober voice of reason is reciting the industry’s preferred frame — safety concern is hysteria, regulation is a “heavy hand,” the companies themselves should decide what to release, anyone who disagrees is participating in a “clamorous moment” that will not matter in retrospect. Capture is when the regulated write the narrative about how well the regulation is going. That is what this column is. Jenkins is not reporting on a frenzy; he is manufacturing its passage — declaring the debate over so his readers can act as if it is.
The line about the moment not mattering? That is the final service the captured voice provides. Telling everyone to stop asking questions.
— Phukher Tarlson