G6: The Hidden Hand

Two joint statements

In December 1953 the president of the American Tobacco Company sent telegrams to the presidents of the seven other major tobacco companies, inviting them to meet. Studies had just linked cigarettes to lung cancer. All but one company came to the Plaza Hotel in New York on 14 December. A few weeks later, on 4 January 1954, the companies published a full-page advertisement in hundreds of newspapers, "A Frank Statement to Cigarette Smokers," announcing a jointly funded research committee and saying that more research was needed. Did they coordinate to mislead the public?

In February 2020, twenty-seven scientists published a joint letter in The Lancet, one of the world's leading medical journals. "We stand together," they wrote, "to strongly condemn conspiracy theories suggesting that COVID-19 does not have a natural origin." Did they coordinate to mislead the public?

Notice which answer came faster, and which way it went. Both describe the same outward event: powerful people agreed and spoke with one voice. Many readers find one suspicious at once and the other obviously innocent, and which one depends less on the evidence given here than on a prior feeling about who is the kind of actor that plots, and who is the kind of person who imagines plots. Those feelings are not evenly spread. In a 2024 Pew survey, 88% of Democrats and 66% of Republicans had confidence in scientists to act in the public's best interests, so the second question lands differently depending on where a reader starts.

The answers, as far as they are known, are below. That feeling is the subject of this page.

What the grammar is

The Hidden Hand is the habit of explaining events by a coordinated actor working in secret. Something large, complicated or painful happens, and the explanation offered is that someone planned it and hid the planning, while the alternatives (many people acting separately on similar incentives, accident, plain bad luck) are set aside.

The deep structure is this: a complex or emergent outcome is attributed to coordinated, concealed agency. ("Emergent" here means an outcome that arises from many separate actions without anyone designing it, the way a traffic jam forms without anyone deciding to cause one.) What makes this grammar unusually dangerous is that it is sometimes right.

The tobacco meeting is one proof. In 2006, after a nine-month trial, a federal judge, Gladys Kessler, found in United States v. Philip Morris that the companies had "known many of these facts for at least 50 years or more" and had "consistently, repeatedly, and with enormous skill and sophistication, denied these facts to the public." The Frank Statement, she found, set out an "open question" position the industry "would maintain for more than forty years." An undated internal Brown & Williamson document, which the court places in the late 1960s, put the strategy in four words: "Doubt is our product."

The CIA's MKUltra program is another: "a closely held CIA project conducted from 1953-1964," as CIA Director Stansfield Turner told a Senate hearing on 3 August 1977, whose tests included giving LSD to "unwitting subjects in social situations." Senator Edward Kennedy told the same hearing that "the records of all these activities were destroyed in January 1973, at the instruction of then CIA Director Richard Helms." More came to light in 1977, after an agency employee whose duties included answering Freedom of Information Act requests searched retired records and found files that earlier searches had missed, because the budget section's retired records had not been searched.

Confirmed cases like these are often invoked to lend credibility to new claims, true or false, that resemble them. Whether they actually change what people believe is less clear; one experiment, described below, found no short-term effect. The confirmed cases are also a skewed sample, as in the toupee fallacy: someone says "every toupee looks fake," never having noticed the good ones. The conspiracies we know about are only the ones that were exposed, and nobody can count the ones that were not.

This site calls a recurring pattern of argument like this a grammar: a shape the same mistake takes across many different subjects, so that once you can see it in one argument you can see it in others. This one has two poles, and each is a real failure.

Pole A, the conspiracists, see coordination everywhere. Their beliefs cannot be falsified, because any evidence against the plot is read as part of the plot. Absence of evidence becomes proof of a cover-up. And they often advance their claims as questions ("I'm just asking questions"), which lets them put an assertion into circulation without ever having to defend it.

Pole B, the debunkers, are so allergic to Pole A's failure that they make the opposite one. They grow complacent about coincidence, assuming that whatever looks coordinated probably wasn't. They use "conspiracy theory" as a label that ends thought rather than as a description that invites checking. And when a claim they lumped in with the cranks turns out to be true, they pay for it with the credibility of every institution that joined in the dismissal.

Taking a hidden-hand claim apart

Neither believing nor dismissing is a method. A few questions do most of the real work.

Who, exactly? A hidden-hand story often names a vague actor: "they," "the elites," "the system," "the deep state." Ask the story to name the people. The tobacco case names the men who received the telegrams. A claim that survives being made specific is much stronger than one that dissolves when you ask for names.

How many people would have to know, and for how long? David Robert Grimes, an Oxford researcher, estimated in 2016 from three exposed cases that "large conspiracies (≥1000 agents) quickly become untenable and prone to failure," while granting that "conspiracies do undoubtedly happen" and survive mainly by "keeping the number of agents low." He acknowledged the skewed-sample problem described above: estimating from exposed conspiracies "might itself introduce bias." But he judged that, "given the highly conservative estimates employed for other parameters," the real chance of a leak was more likely to be higher than his estimate, not lower. The independent scholar Kurtis Hagen replied in 2022 that "at least some types of conspiracies do not tend to fail" (only his summary has been read for this page). The question is worth asking; its answer is a probability, not a verdict.

What would we expect to see if nobody coordinated? Many outcomes that look planned come from people with similar incentives acting separately. If every company in an industry faces the same rising costs, their prices may rise together without a meeting. Ask whether the pattern is better explained by a plan or by shared pressures.

Is the coordination secret at all? Both statements in the opening were public and signed. What the tobacco companies concealed was not that they had met but what their own research showed. A claim about open coordination is a claim to be checked against the public record, not a theory about concealment. Part of a coordination can be open and part hidden, as the second statement turns out to show.

How would it be exposed, and by whom? The confirmed cases on this page came apart through slow, ordinary institutions. The tobacco case was proved in court, after the companies were made to hand over internal documents in discovery (the pre-trial process in which each side must produce relevant records). MKUltra surfaced through a Senate inquiry and an agency records search. Watergate began to come apart as Washington Post reporters tied the scandal to top officials in the Nixon administration. Ask of any claim who could expose it, and whether anyone is trying. A story in which every possible exposer is part of the plot has stopped being checkable.

What would count against the claim? If no possible evidence could show the coordination did not happen, the claim is not really about the world. This applies to the dismissal too: if no possible evidence could convince you that a particular institution did coordinate, your certainty is not about the world either.

What is the track record? Some institutions have a documented history of concealment. That history is evidence and should raise suspicion. It is not proof that this particular claim is true.

Why it is so persistent

Psychologists call the capacity to recognize intentions behind events agency detection. In a 2018 review, Jan-Willem van Prooijen and Mark van Vugt report studies linking conspiracy belief to a tendency "to overascribe intentionality to inanimate objects," and argue that suspicion of plots may have evolved because, in ancestral groups, missing a real hostile coalition could be deadly. They call their support "tentative." It is a plausible story, not a settled finding.

A plan can also be more satisfying than an accident. Reviewing the research in 2017, Karen Douglas, Robbie Sutton and Aleksandra Cichocka group the motives behind conspiracy belief as "epistemic (understanding one's environment), existential (being safe and in control of one's environment), and social (maintaining a positive image of the self and the social group)," and find that "conspiracy belief may be more appealing than satisfying." They are careful about the other pole too: "history has repeatedly shown that corporate and political elites do conspire against public interests."

This page suggests that the pole that dismisses conspiracies has comforts of its own: it lets a person trust familiar institutions without auditing them, and marks the dismisser as sensible. No study read for this page measures those comforts, so take this as an argument, not a finding. The blogger Scott Alexander, in "Epistemic Learned Helplessness", describes deciding that in some fields "any attempt to evaluate the arguments is just going to be a bad idea so I don't even try," and defends it as "a pretty useful social safety valve" for people who know they can be argued into anything. His own use of it is to side with established experts over persuasive outsiders. This page suggests the stance can be turned the other way, by a conspiracist who declines to weigh official evidence. Used everywhere, in either direction, the safety valve becomes a way of never being moved.

Real scandals also feed distrust. In Pew Research Center's long-running measure, 73% of Americans trusted the federal government to do the right thing almost always or most of the time in 1958; trust eroded with Vietnam and Watergate, and stood at 17% in September 2025. Whether a real scandal makes people believe unrelated conspiracy claims is less clear: in a 2023 experiment with 483 students, reminders of real scandals "did not affect their conspiracy beliefs," at least in the short term, and stronger believers were "slightly better" at recognizing invented scandals but "somewhat worse at recognizing real scandals" (only the abstract and a press summary could be read). One small study settles nothing, but it cuts against the idea that conspiracy belief is simply scandal-literacy.

How the argument has been had before

Illuminati and Masonic panics. As the historian Richard Hofstadter told it in "The Paranoid Style in American Politics" (1964), the Bavarian Illuminati was real, founded in 1776. In a book that reached America in 1797, the Scottish scientist John Robison claimed, as Hofstadter quotes him, that it aimed at "overturning all the existing governments of Europe," and credited it with the French Revolution; New England pulpits took up the alarm. Of the anti-Masonic movement of the 1820s and 1830s Hofstadter wrote, "Certain elements of truth and reality there may have been." The paranoia was the leap from a secret society existing to its controlling everything. He insisted the style was "not necessarily right-wing," naming "certain spokesmen of abolitionism" and "the popular left-wing press" among its users.

The forged Protocols of the Elders of Zion. According to the United States Holocaust Memorial Museum, the Protocols appeared in Russia in 1903 as the supposed minutes of secret Jewish meetings, and in 1921 The Times of London showed that much of it was copied from an 1864 French satire that does not mention Jews. It circulated anyway, through Henry Ford's The International Jew and Nazi propaganda. Joseph Goebbels wrote that he believed it "is a forgery," and then: "I believe in the inner, but not the factual, truth of The Protocols." That is the hidden hand at its worst: an invented coordinator, a hated group, and a belief beyond the reach of evidence. Debunking alone did not stop it.

The assassination of John F. Kennedy. The Warren Commission concluded in 1964 that "Oswald acted alone." In 1976 a Senate committee found that "neither the CIA nor the FBI told the Warren Commission about the CIA attempts to assassinate Fidel Castro." In 1979 a House committee, relying largely on acoustic evidence, found Kennedy "was probably assassinated as a result of a conspiracy"; in 1982 a National Research Council panel found the sounds were recorded "about one minute after the President had been shot" and do not support "a second gunman." Each pole was partly right: agencies did withhold relevant information, and the second-gunman evidence fell apart. Gallup found that 52% thought others were involved in the week of the shooting, 81% in 1976, the year the Senate committee reported, and 61% in 2013.

9/11 trutherism. After the attacks of September 11, 2001, a movement grew claiming the US government had planned or knowingly allowed them, often holding that explosives brought the towers down. The 9/11 Commission documented real failure ("The most important failure was one of imagination") and missed chances, such as "not watchlisting future hijackers Hazmi and Mihdhar." The federal engineering investigation found "no corroborating evidence" for controlled demolition. The truthers took documented incompetence and read it as intent.

QAnon. QAnon emerged in 2017 on anonymous message boards, promising a day of mass arrests of its enemies. When Joe Biden was inaugurated on 20 January 2021 instead, the Associated Press reported that some believers decided "Biden's victory was an illusion" or that the ceremony was "computer-generated," while one prominent promoter, Ron Watkins, wrote on Telegram, "We gave it our all," and then, "Now we need to keep our chins up and go back to our lives as best we are able." Both responses to a failed prediction were on display: absorbing it into the plot, and letting it count.

The other pole's history is harder to find. Two days after the Watergate break-in of 17 June 1972, President Nixon's press secretary, Ron Ziegler, said that "certain elements may try to stretch this beyond what it is"; he became known for calling it a "third-rate burglary," and he dismissed the Washington Post's reporting as inaccurate. In 1973 he apologized to the paper: "When we are wrong, we are wrong, as we were in that case." The tobacco industry's "open question" was the same kind of move. But these were the conspirators' own cover stories, part of the hidden hand itself, not outsiders reasoning badly. The debunker's failure is something else: a journalist, official or scientist who was not in on it, waving away a claim that later proved true. This page has not yet found a quoted, dated case of that kind for MKUltra, Watergate or tobacco. That is a gap in the research so far, not evidence that the failure is rare.

How each pole fails

The conspiracist starts from the conclusion that someone is behind it, and reads every fact as confirmation. Evidence against the plot is a cover-up; missing evidence is a cover-up too. The failure is not limited to cranks. Hofstadter wrote of Robison, a respected scientist, that he "seems to have made his work as factual as he could," but made "the characteristic paranoid leap into fantasy" when he came to judge the society's influence. The debunker starts from the conclusion that no one is behind it, and the label "conspiracy theory" arrives before the evidence is weighed.

Which pole a person stands at often depends on who is accused. At a press conference in Helsinki on 16 July 2018, asked whether he believed his intelligence agencies, which had concluded that Russia interfered in the 2016 election, or President Putin, who denied it, President Trump asked about a missing Democratic Party server ("Where is the server?"; "all I can do is ask the question") and said of Russia: "I don't see any reason why it would be." The next day he said he had meant "wouldn't." The Mueller report (2019) found that Russia interfered "in sweeping and systematic fashion." It also "did not establish" that the Trump campaign "conspired or coordinated" with Russia, and it warned against reading that too strongly: "A statement that the investigation did not establish particular facts does not mean there was no evidence of those facts." The server question has the shape of Pole A. The waving-off of the interference finding is harder to classify: it may be the debunker's reflex, or simply taking the accused's word, and this page does not count it as an example of Pole B. In 2026, after claims spread that a shooting at the White House Correspondents' Dinner had been staged, the same speaker took the other side, telling 60 Minutes, as NPR reported: "I think they're more sick than they are con people."

The scientists' letter from the opening is this page's present-day example of the debunker's failure, and the opening's question about it has a partial answer. Emails later released under public-records requests, and published by U.S. Right to Know, an advocacy group that campaigns on COVID-origin questions, show that one of its signers, Peter Daszak, president of the non-profit EcoHealth Alliance, organized it: he edited the draft with co-signers' comments and planned how to gather more signatures. (U.S. Right to Know says he drafted it.) In an email dated 6 February 2020 by its file name, he wrote that the statement "will not have EcoHealth Alliance logo on it and will not be identifiable as coming from any one organization or person, the idea is to have this as a community supporting our colleagues." So the letter was coordinated, and readers could not see who had coordinated it. Whether anyone meant to mislead is a different question, and nothing read for this page answers it. The signers declared "no competing interests." In June 2021 The Lancet published an expanded disclosure from Daszak, stating that his organization's work in China had been funded by the US National Institutes of Health and included "the production of a small number of recombinant bat coronaviruses." The journal noted that "There may be differences in opinion as to what constitutes a competing interest."

Was the letter's labelling a failure? The charitable reading is that in early 2020 some origin stories really were conspiracy theories. The same email discusses "the bioengineered virus conspiracy theory" and says "there are other conspiracy theories out there." The letter also gave its grounds: genome analyses that "overwhelmingly conclude that this coronavirus originated in wildlife," and a letter from the presidents of the US National Academies. In this page's reading, the charitable account explains the letter's purpose but not its wording. To a co-signer who wanted the bioengineering claim addressed directly, Daszak wrote: "you're right it would be good to be specific about the bioengineered virus conspiracy theory, but we I think we should probably stick to a broad statement" (the slip is in the original). The broad wording condemned every suggestion that the virus "does not have a natural origin." In January 2025 the CIA judged a lab origin more likely, with "low confidence," while scientists, the Associated Press noted, think a route from bats through a Wuhan market most likely. As of October 2026 the origin is unresolved. This page's reading is that the letter's failure was not its conclusion, which may yet prove right, but a label broad enough to cover a question that was, and remains, open.

Conspiracism is not confined to one party either. In a NewsGuard/YouGov poll reported by NPR on 11 May 2026, 42% of Democrats said the July 2024 attempt on Trump's life in Butler, Pennsylvania, was staged, against 7% of Republicans, despite "numerous witnesses" and a Senate investigation. The picture is not static. About the dinner shooting a few weeks before the poll, the share of Republicans calling it staged rose to 13%, against 34% of Democrats. A NewsGuard editor, describing the staging claims posted online after the dinner, said: "There's really not a lot of evidence that these social media users are citing or relying on." NewsGuard is a company that rates news sources and tracks false claims online. It is itself contested: as the Associated Press reported in March 2026, the Federal Trade Commission "accuses the company of trying to suppress conservative speech," and a conservative group has published studies arguing its ratings favour liberal outlets, while NewsGuard says its ratings follow clearly defined criteria. This page has so far found the left's share of Pole A in polling rather than in a named public figure's own words. That is a limit of the research, not a finding that the left's version is milder.

Both poles share one move. Each decides in advance what kind of actor plots and what kind of person imagines plots, and then lets that decision stand in for the evidence. Most people, if they are honest, hold both positions: conspiracist about the institutions they distrust, debunker about the ones they trust.

Where it borders other grammars

This grammar rarely travels alone.

With G12: Fixed Pie or Growing Pie it shares the search for a thief. If every gain must have come at someone's expense, every success needs a culprit, and the hidden hand supplies one.

With G7: Moral Panic and the Folk Devil it shares a villain. A moral panic blames a visible group for a social problem; a hidden-hand story often blames a concealed one, and the two merge when the folk devil is said to be secretly organized, as in the Protocols.

With G11: Contested Epistemic Authority it shares the question of whom to trust. Whether a claim of coordination is believable often depends less on the evidence than on whether the reader trusts the institutions accused, and the people who deny it.

When naming the pattern becomes a weapon

"Conspiracy theory" is a very effective phrase for ending a discussion without having one. Once a claim has been given that label, the person making it has to defend not just the claim but their own sanity. That is useful when the claim is baseless. It is a weapon when the claim is serious and the label is applied because the claim is inconvenient, or, as this page reads the Lancet letter, when the label is drawn so broadly that it covers an open question along with the baseless ones. It is the weaponized form of D4: The JAQ Detector (JAQ is short for "just asking questions"), the tool for spotting claims smuggled inside questions: institutions labelling any hard, fair question a conspiracy theory to avoid answering it.

The weapon points the other way too. The conspiracist has a vocabulary of dismissal too ("shill," "sheep"), and conspiracy theories, as Douglas and her colleagues note, imply "that people who try to debunk conspiracy theories may, themselves, be part of the conspiracy." Robert Welch, founder of the John Birch Society, called President Eisenhower, as Hofstadter quotes him, "a dedicated, conscious agent of the Communist conspiracy." The broadcaster Alex Jones called the 2012 Sandy Hook school shooting a hoax and, as Columbia University's Global Freedom of Expression project (a university center that promotes free expression) summarizes a 2022 Connecticut ruling, questioned "the sincerity of the reactions" of bereaved families. The families' claims included defamation. In October 2022 a jury awarded them damages, and in the ruling the summary describes, the main issue was the amount Jones owed. Those with the strongest evidence against the plot were folded into it.

This page can be turned into a weapon too. Its vocabulary ("unfalsifiable," "agency detection") could be used to dismiss anyone who suspects coordination, which is Pole B with a glossary; its confirmed cases could be used to excuse any claim at all, which is Pole A with a footnote. A label that names a real failure in general does not settle whether the failure is happening here.

Catching it in yourself

Go back to the two joint statements. Which did you suspect first? If it was the one whose signers you already distrusted, you have found the grammar in yourself. Then notice how you took the answers: the tobacco companies did coordinate to mislead, and the scientists' letter was coordinated out of sight, for purposes that nothing read for this page shows to be dishonest. Did either answer feel like a win for your side?

Which institutions do I assume would lie, and which do I assume wouldn't? Make two lists. If the first list is your political opponents' institutions and the second is your own side's, the lists are doing your thinking.

When did I last call something a conspiracy theory, and did I check it first? Or did the label arrive before the evidence?

When did I last suspect coordination, and what would have convinced me I was wrong? If the answer is nothing, the belief is not about the world.

Have I ever been wrong in this direction? Can you name a claim you dismissed that turned out true, or a plot you suspected that turned out to be coincidence? If you can't name one of each, ask whether that is because you have never been wrong, or because you have never checked.

Would I accept this explanation if it blamed my own side? Swap the accused for an institution you trust, and see whether the evidence still feels like enough.

Tools that help

• C1: The Falsifier Question. Write down, in advance, what observation would count against the claim. A hidden-hand claim that can absorb any evidence fails this test, and so does a dismissal that no evidence could overturn. Its weaponized form is demanding falsifiers from the other side while offering none of your own.
• A2: The Denominator Demand. Ask out of how many. Ten suspicious coincidences sound damning until you ask how many events were searched to find them, and how many coincidences of that kind would turn up by chance. Its weaponized form is drowning a real, concentrated pattern in a huge base rate (the background frequency of something across all cases) to make it disappear.
• D4: The JAQ Detector. Watch what happens when a question receives an answer. A genuine questioner updates; someone using the question to carry a claim moves on and asks it again somewhere else. Turned the other way, it catches institutions that call a fair question a conspiracy theory to avoid answering it.
• C5: The Cynicism Firewall. Skepticism grades claims one at a time; cynicism rejects everything and calls it wisdom. It is the tool for the person who has learned of one real scandal and is tempted to distrust everything, and for the person who has seen many false theories and is tempted to dismiss every unofficial claim without looking at it.

Where it shows up now

As of October 2026, the site has not yet examined this grammar in a live argument. The following are places to look, not findings.

• Jeffrey Epstein. Any examination will keep established facts (his 2008 guilty plea; his death in federal custody in August 2019, ruled a suicide by officials) separate from allegations about living people, which are not repeated, and will not adjudicate the cause of death. In February 2026, 75% of Americans told Reuters/Ipsos the government is hiding information about his alleged clients, after a Justice Department and FBI review, made public on 7 July 2025, said it found "no incriminating 'client list.'" In the same poll, majorities of Democrats (81%), independents (68%) and Republicans (58%) agreed that the files show the powerful "are rarely held accountable for their actions." Not yet examined.
• "15-minute cities," an urban-planning idea for putting daily needs within a short walk or ride. In Oxford in late 2022 and early 2023, as the student newspaper Cherwell reported on 14 January 2023, an online article falsely claimed that new traffic filters were a "climate lockdown" that would lock residents into zones, and councillors received death threats; the same report notes "legitimate opposition" to the filters, including petitions. Not yet examined.
• Claims about election machinery. Not yet examined.
• "The algorithm is suppressing me." Not yet examined.
• Distrust of the pharmaceutical industry. Not yet examined.
• "AI companies know something." Not yet examined.

Related concepts

• Base rates of successful conspiracies. How often secret coordination of a given size actually happens and stays hidden.
• Agency detection. The capacity to perceive an intentional actor behind events, which can misfire when events have no author.
• Emergent outcomes. Results that arise from many separate actions without anyone designing them.
• Unfalsifiability. The property of a claim that no possible observation could count against.
• Toupee fallacy. Judging a whole category from only the members you noticed, when the ones you didn't notice would change the verdict.
• "Just asking questions." Putting a claim into circulation inside a question so that the asker never has to defend it.
• Thought-terminating label. A word or phrase that ends consideration of a claim instead of advancing it.
• The paranoid style. Richard Hofstadter's 1964 name for "heated exaggeration, suspiciousness, and conspiratorial fantasy" in politics.
• Epistemic learned helplessness. Scott Alexander's term for declining to evaluate arguments in a field where you know you can be talked into anything, which he defends as often sensible.
• Metastasized distrust. The spread of justified suspicion about one institution's conduct into distrust of everything it says.