AI Safety Panic Is 'Epiphenomena' — Whose Bubble Is Jenkins Really Defending?

Responding to: Unintelligence on Super Intelligence — Holman W. Jenkins, Jr. · 2026-10-09

What the Piece Argues

Holman Jenkins argues that recent public panic over AI superintelligence is “epiphenomena” — a side-show obscuring the actual AI story, which is that frontier labs (Anthropic, OpenAI) already operate under de facto intelligence-agency control via the NSA and a “voluntary” Trump-administration process he concedes is not really voluntary. He contends this existing entanglement — a former four-star NSA chief on OpenAI’s board, Anthropic distributing its top models “in collaboration with the US government,” the Defense Production Act of 1950 ready to be invoked — renders new AI regulation unnecessary, and he dismisses AI safety critics like David Robinson as former employees doing post-departure lamentation in public. He folds in a related claim that any market turn against Elon Musk carries parallel SEC and criminal-prosecution risk, framing the AI debate as a way station on the road to a Musk reckoning. The piece functions as a defense of frontier AI’s existing governance arrangements and a delegitimization of public safety concerns as theatrical.

Receipts

Jenkins’s column hands you, in its own prose, every receipt that defeats its own conclusion — and then asks you to read past them.

The framing wants you to believe

• The recent surge in AI safety concern is panic-driven theater unrelated to the real story.
• Frontier labs are already subject to effective oversight via the NSA and a Trump-administration “voluntary” process.
• New AI legislation is unnecessary; the existing arrangement (intelligence-agency entanglement) is working.
• Critics like David Robinson are disgruntled former employees touring the press.
• Anthropic’s “in collaboration with the US government” model-distribution is normal, not alarming.
• The “super intelligence” everyone worries about exists only as “a rumor from the deepest bowels of the world’s top intelligence establishments.”

What’s really going on

• The piece concedes in print that a former four-star NSA chief sits on OpenAI’s board — the national-security state has formal representation in the corporate governance of the leading AI lab.
• It concedes Anthropic this week allowed certain customers access to advanced models only “in collaboration with the US government” — frontier-model distribution is already subject to executive-branch gatekeeping.
• It concedes the “voluntary” Trump process “isn’t really voluntary” and that the Defense Production Act of 1950 is “ready to be invoked as necessary” — the executive branch holds dormant authority to commandeer the industry at will.
• It concedes the Altman/Brockman 2017 debate over pre-payment to China, Russia, and the U.S. for tools they expected to be expropriated — i.e., lab founders themselves expect the state to capture their work.
• The piece concedes all of this and then turns around and tells readers the public’s safety concerns are “epiphenomena” — i.e., the real story is that AI is being governed by the security state, and the public discussion of safety is a distraction.
• Anchor: Jenkins, “Unintelligence on Super Intelligence,” Wall Street Journal Opinion, October 9, 2026 (print edition under the same headline). The concessions above are the column’s own prose; the framing the piece advances is undermined by its own receipts.

The Response Ladder

Polite Reframe

When to use: a friend who sends you the column and wants your honest read; a small-business owner or working professional whose work is being reshaped by these models; any persuadable reader who can be reached with a calm, receipts-anchored moral reminder.

A parent in a public-school district used an AI tutoring tool last spring that the vendor said was “aligned” with appropriate-use standards and that no data would leave the classroom. Then someone read the small print. The data — every reading assessment, every conversation, every flag for the kids struggling with phonics — flowed through a model whose distribution rights sat inside a corporation whose board includes a retired four-star general from the National Security Agency. The general’s previous life was cyber and signals intelligence. The data wasn’t going to a vendor server. It was going to a system entangled, by the company’s own admission, with U.S. national-security apparatus. The parent stopped using the tool.

I tell you about this because the editorial page of The Wall Street Journal this month ran a piece by Holman Jenkins telling readers that public concern about AI safety is “epiphenomena” — a side-show, not the main event. The main event, in Jenkins’s telling, is that AI is already subject to “a de facto regulator” in the NSA, via a Trump-administration “voluntary” process that Jenkins himself concedes “isn’t really voluntary.” Jenkins argues this existing arrangement — frontier AI labs operating in collaboration with the U.S. intelligence community — is fine, and that the public conversation about AI safety is a distraction.

Read the column. The receipts are in the prose. Jenkins tells us a former four-star NSA chief sits on OpenAI’s board. Jenkins tells us Anthropic this week allowed certain customers access to advanced models only “in collaboration with the US government.” Jenkins tells us the Defense Production Act of 1950 is “ready to be invoked as necessary.” Jenkins tells us the labs’ own founders, in 2017, debated how to get paid up front for tools they expected the government to “expropriate.” Jenkins does all this, and then turns around and tells us the public’s safety concerns are theater.

The question isn’t whether the government is involved in frontier AI. The government is involved, deeply, by Jenkins’s own admission. The question is who decides how that involvement is structured — a public regulatory process with hearings, votes, and accountability, or an intelligence-agency arrangement negotiated behind closed doors with company boards, where the public learns the rules only when a model gets withheld, released, or steered.

A senator who takes direction from a foreign intelligence service and votes that service’s bill is, by long conservative tradition, a foreign agent. The corollary holds. A company that takes direction from a domestic intelligence agency and produces the technology the agency wants — with a former four-star general on the board to keep the arrangement tidy — is an agent of a particular kind. Get the intelligence apparatus out of corporate governance, or stop pretending we have a public AI policy. Either is fine. Pick one.

If you want to do something useful with the column this week, call your representative’s office and ask whether the AI governance framework being built right now is going through Congress or through the NSA. If they don’t know, that is the column’s most important sentence — and Jenkins didn’t write it.

Mockery and Ridicule

When to use: an investor chat, a tech-policy mailing list, a Substack comment section where the “epiphenomena” line is being repeated approvingly; any forum where bystanders outnumber believers.

So: the Wall Street Journal editorial page — the institutional voice of free markets, small government, and the wisdom of the private sector — has discovered that the most efficient regulator in the American system is the National Security Agency. Hooray. Forty years of WSJ op-eds warning that regulators capture the industries they oversee, and the editors find peace at last when the regulator in question is the one with the spy satellites.

Jenkins’s argument, once you strip the jargon, is: stop worrying about AI safety, because the U.S. intelligence community already has the keys. A former four-star NSA chief sits on OpenAI’s board. Anthropic distributes its top models only “in collaboration with the US government.” The Trump administration has set up a “voluntary” process — Jenkins’s own word for it, and he immediately admits “it isn’t really voluntary” — that hands the frontier labs a “de facto regulator” in the NSA. The Defense Production Act of 1950 is “ready to be invoked.” The lab founders themselves, in 2017, sat around debating how to get paid up front by the governments they expected to expropriate their work. And yet — read the column — the public’s safety concerns are “epiphenomena.” The “super intelligence” everyone worries about, we are told, “you will likely encounter… only as a rumor from the deepest bowels of the world’s top intelligence establishments.”

Imagine defending this arrangement with a straight face. The same editorial page that spent a generation teaching its readers that the FDA captures pharma, that the FCC captures telecom, that the SEC captures finance, that the CFPB captures banks — that page now assures us that when the NSA captures the most transformative industry of the century, with no hearings, no votes, no public rule-making, and a former four-star general sitting on the corporate board of the leading lab, that this is just fine, and the people worried about it are theatrical.

It is not theatrical. It is the most basic question of democratic governance there is. Who decides what the most powerful technology ever built does, on whose authority, with what accountability to the public whose lives it will touch? Jenkins’s answer: a former NSA director on OpenAI’s board, an Anthropic press release, a “voluntary” Trump process, and a 1950 wartime statute ready to be invoked. The public’s only role, apparently, is to relax.

Imagine, if you can stomach it, the funeral procession the WSJ editorial page would lead if a Democratic administration put a single former intelligence official on the board of a major tech company. Imagine the columns about regulatory capture. Imagine the editorial cartoons. Imagine the hearings Republicans would demand. Now imagine the same editorial board discovering that the same arrangement, but executed by their own side, is just fine, and that the public’s safety concerns are theater.

The “epiphenomena” line is doing a lot of work here. It is the editorial-page equivalent of “calm down, dear, it’s only a coup.” When a piece of that opinion earns the headline Unintelligence on Super Intelligence, the editorial board is telling you, by accident, exactly what they think of their own argument.

If you want to use the column for anything, use the receipts Jenkins hands you. He has just confirmed, in print on the page of America’s paper of record, that the U.S. intelligence community is the de facto regulator of frontier AI. You don’t have to take anyone’s word for it. The WSJ editorial board has done it for them. That is the most useful thing the column has produced this year, and I doubt it was the column’s intention.

Nuclear Satire

When to use: a long-form essay in a left-leaning policy publication; an extended thread for tech-policy readers who appreciate baroque political satire; any venue where the grotesque metaphor of intelligence-agency capture can land at full force.

Let the record show: the editorial board of The Wall Street Journal — the institutional voice of free enterprise, limited government, and the proposition that the private sector regulates itself more wisely than the state ever could — has, in October of the year 2026, published an op-ed arguing that the appropriate regulator of artificial intelligence is the United States National Security Agency. Not the courts. Not Congress. Not a federal agency with administrative procedure, public hearings, rule-making, and judicial review. The NSA. The agency that for sixty years has operated behind the deepest classification walls in the American government, accountable to no court in the ordinary sense, FOIA-exempt in practice, with a budget the Director of National Intelligence himself does not fully disclose to the public. That NSA. The Jenkins column is on the page. You can go read it.

The arrangement, as Jenkins describes it with the unblinking air of a man describing the weather: a “voluntary” Trump-administration process that Jenkins immediately concedes “isn’t really voluntary” hands the frontier AI labs a “de facto regulator” in the NSA. A former four-star NSA chief sits on OpenAI’s board. Anthropic releases its top-tier models only “in collaboration with the US government.” The Defense Production Act of 1950 — passed to mobilize industry for the Korean War, never repealed, available for use at the President’s discretion in pursuit of “national-security goals” — is, in Jenkins’s word, “ready to be invoked as necessary.” The lab founders themselves, in 2017, debated how to get paid up front by China, Russia, and the U.S. for tools they expected governments to expropriate. We are told all of this in calm, prose-of-the-page, business-section diction. The diction is the cover-up. The diction is the point.

Imagine the funeral procession the WSJ editorial page would lead if a Democratic administration put a single former intelligence official on the board of a major tech company. Imagine the columns about regulatory capture. Imagine the hearings. Imagine the editorials about the deep state. Now imagine the same editorial board discovering that the same arrangement, but executed by their own side, is just fine, and that the public’s safety concerns are “epiphenomena.” Now imagine they put it on the page under the headline Unintelligence on Super Intelligence, in the print edition, where the editors are not even pretending it was an oversight.

The “epiphenomena” line is a small masterpiece of editorial-page diction. It is the in-house term of art for “I am going to ask you to believe that the thing you are watching happen is not the thing that is happening.” The actual, primary phenomenon — the phenomenon any honest editorial page would name first — is that a transformative technology is being governed, without statute, without hearing, without vote, by an intelligence agency that has spent sixty years perfecting the art of governing things without anyone noticing. The “epiphenomena” is the editorial-board euphemism for this. The “rumor from the deepest bowels of the world’s top intelligence establishments” is the editorial-board romance of the same arrangement. The “draconian Defense Production Act of 1950 ready to be invoked” is the editorial-board promise that if the voluntary arrangement fails, the involuntary one is already loaded.

Let us be precise about what is on the page. The WSJ editorial board, in October 2026, has told its readers: do not worry about how the AI industry is governed. The arrangement is fine. The arrangement is the intelligence community. The arrangement is the executive branch commandeering a transformative industry through a process that does not pretend to be a process. The arrangement is the kind of thing the editorial board would, in any other context, call a constitutional emergency.

Jenkins tells us the “super intelligence” you will likely encounter is “a rumor from the deepest bowels” of the intelligence community. I have read that sentence three times. I cannot make it mean anything other than what it means: that the editorial board has made its peace with intelligence-community monopoly over the most consequential technology of the century, and would like its readers to do the same. The headline of the print edition is Unintelligence on Super Intelligence. The editorial board knows what they did. They are not sorry about it. They are bragging.

Profane Scorched-Earth

When to use: the polite reframe has been sent; the column is being passed around; the audience needs the full cathartic register at the profane apex; the moment when anger done well is the work.

What the fuck does Holman Jenkins think he’s doing.

I want to be clear about what is on the page of The Wall Street Journal’s editorial section this week, because the prose disguises itself and the headline makes jokes. The piece is called “Unintelligence on Super Intelligence.” Holman Jenkins is a member of the editorial board of The Wall Street Journal. The Wall Street Journal editorial page is, in 2026, the most powerful institution in American conservative intellectual life. The op-ed is titled in the print edition Unintelligence on Super Intelligence, which is the editorial board admitting, in the space above the table of contents, that they know exactly what the fuck they are doing.

The piece argues — and I am going to type this slowly because the diction hides it — that the appropriate regulator of artificial intelligence is the United States National Security Agency. Not the courts. Not Congress. Not an agency with administrative procedure. The NSA. An institution that has spent sixty years operating behind the deepest classification walls in the American government, accountable to no court in the ordinary sense, FOIA-exempt in practice, whose budget the DNI himself does not fully disclose. That NSA. The piece is on the page. You can go read the goddamn thing.

And here is the part that should make you spit. Jenkins, in print, concedes every fucking receipt that makes this argument indefensible, and then turns around and tells you the public safety concerns about it are “epiphenomena.” His own words. He concedes a “voluntary” Trump process that “isn’t really voluntary.” He concedes a former four-star NSA chief sitting on OpenAI’s board. He concedes Anthropic distributes its top models only “in collaboration with the US government.” He concedes the Defense Production Act of 1950 is “ready to be invoked as necessary.” He concedes the lab founders in 2017 sat around debating how to get paid up front by the governments they expected to expropriate their work. He concedes that a San Francisco conference executive said there is “some level of machine intelligence beyond which the industry should be prepared not to go” — and then notes, with what I swear is a smirk, “we’ll see how the NSA feels about that.”

He concedes all of this in print on the editorial page of America’s paper of record, and his conclusion is that the public panic about AI is theater. His epiphenomena. His little side-show. His nothing-burger. While the NSA is being handed the keys.

This is where the anger needs to land — and not at the American voter who has been told to mistrust regulation by thirty years of WSJ editorial-page work. The American voter was told the FDA captures pharma. The American voter was told the FCC captures telecom. The American voter was told the SEC captures finance. The American voter was told the deep state was an existential threat to American liberty. The American voter believes that. The American voter was told to believe that. They are not the enemy here. They were used. They are being used right now, in this column, in this prose, in this specific fucking paragraph.

The enemy is the editorial page that told them. The enemy is the arrangement that the editorial page is now defending. The enemy is the WSJ’s editorial board discovering, in October of 2026, that the deep state they spent a generation warning you about is actually just fine when it is deep state for their side, running the industry they want protected, governing without statute, without hearing, without vote, with a retired four-star general sitting on the goddamn board of the leading lab to keep the arrangement tidy.

You want a canon, you want a citation? Here is one. Jeremiah, eighth chapter, twelfth verse: “Were they ashamed when they had committed abomination? nay, they were not at all ashamed, neither could they blush.” The prophets had a name for people who had lost the capacity to be ashamed of what they were doing. The prophets did not think much of them. I do not think much of the editorial page that publishes an op-ed telling its readers that intelligence-agency monopoly over the most consequential technology of the century is “epiphenomena” theater, and that the public should relax about it, and that the print edition headline should be a fucking pun about it. They knew not how to blush. They are not blushing now.

The “rumor from the deepest bowels of the world’s top intelligence establishments” — Jenkins’s own phrase, in the prose, in the piece, in the print edition. He wants you to meet the most consequential technology of the century as a rumor. He wants you to take its governance on faith. He wants you to trust that the arrangement he has just described in print, in detail, with the receipts in his own sentences, is being overseen by people whose institutional habit, for sixty years, has been to oversee things without telling you what the fuck they are doing.

I am not going to be polite about this. The polite reframe is up the page. The polite reframe is for your Thanksgiving uncle. The polite reframe is for the friend who sent you the column and wants to be talked off the edge. That work has been done. For the rest of you, the editorial board of The Wall Street Journal has just confirmed, in print, on the page, in prose you can copy and quote, that the U.S. intelligence community is the de facto regulator of frontier artificial intelligence. You don’t have to take my word for it. You don’t have to take the industry’s word for it. You don’t have to take the AI safety advocates’ word for it. The Wall Street Journal editorial page has done it for them, in their own voice, with their own receipts, and the headline of their own print edition is “Unintelligence on Super Intelligence” because they are not even pretending it was an oversight.

So what the fuck does Holman Jenkins think he is doing. He thinks he is doing what the Wall Street Journal editorial page has always done: defending the arrangement that protects the industry that pays for the page. He just did it with the NSA this time, instead of with the FDA or the FCC or the SEC. The institution is the same. The argument is the same. The conclusion is the same. The only thing that changed is which arm of the federal government is doing the regulating, and the editorial page is fine with it, because the editorial page was never actually against the regulating. The editorial page was against the regulating when the regulating was on the other side.

That is what “epiphenomena” means. It means: please do not look at this.

Don’t you fucking dare look away.