Game theory cannot tell you that a conspiracy happened. What it can do is test whether the conspiracy being alleged makes strategic sense.
If a hypothesis requires several people to secretly coordinate, game theory gives us a structured way to ask who those people are, what each person gains by cooperating, what they risk by defecting, whether secrecy is stable, and what behavior the arrangement should produce.
That matters because two bad shortcuts dominate discussions of conspiracy theories.
One is:
There are unanswered questions, therefore there was a conspiracy.
The other is:
That’s a conspiracy theory, therefore it can be dismissed.
Neither follows logically.
The better sequence is:
observation → competing hypotheses → strategic plausibility → predicted evidence → comparison with what we actually observe.
Game theory belongs in the middle of that process. It cannot manufacture evidence. But it can expose a hypothesis that quietly depends on irrational actors, impossible coordination or nonexistent incentives. It can also show when a supposedly implausible conspiracy may actually have mechanisms that make cooperation and secrecy rational.
A conspiracy label does not settle the evidence
Even the terminology is more complicated than it first appears.
The Internet Encyclopedia of Philosophy’s review of conspiracy theories notes that philosophers have long disagreed over whether conspiracy theories should be treated as an inherently defective category or whether individual conspiracy explanations should be evaluated on their merits.
A 2026 paper by Romy Jaster and Geert Keil in Erkenntnis tries to reconcile those positions. The authors argue that conspiracy theories may exhibit recurring epistemic problems while still maintaining that particular conspiracy hypotheses deserve case-by-case investigation.
That distinction is useful.
For purposes of evaluating evidence, conspiracy hypothesis is often the cleaner term. It means a proposed explanation in which people secretly coordinated to produce, exploit or conceal an outcome. Whether that hypothesis is well supported is a separate question.
This is not merely philosophical housekeeping.
A 2026 British Journal of Social Psychology study examined people’s ability to distinguish documented, verified conspiracies from unverified conspiracy claims. Across three studies, better discrimination was associated with measures including general knowledge and scientific reasoning. The researchers found that accurately distinguishing between the categories was more informative about cognitive sophistication than simply having a tendency to accept or reject conspiracy claims.
The authors also make an important qualification: “verified” and “unverified” are proxies for evidentiary status, not perfect access to truth. Some genuine conspiracies may remain undiscovered, while an unverified claim could later acquire evidence.
The lesson is not that conspiracy theories deserve a presumption of truth.
It is that good reasoning requires discrimination rather than reflex.
Unanswered questions are a starting point, not proof of coordination
Suppose a major event contains three genuinely puzzling facts.
Those anomalies deserve investigation. But they do not, by themselves, tell us what caused them.
Several explanations might fit:
| Hypothesis | Possible explanation |
|---|---|
| H1 | Ordinary error or incompetence |
| H2 | Coincidence |
| H3 | Several actors independently responding to the same incentives |
| H4 | Misconduct by one actor |
| H5 | A small group secretly coordinating |
| H6 | A broader coordinated conspiracy |
| H7 | Some combination of the above |
This is where conspiracy reasoning frequently goes wrong.
An unexplained fact becomes evidence that something remains unexplained. It is not automatically evidence for whichever hidden explanation is most interesting.
A serious conspiracy hypothesis therefore needs more than anomalies. It needs a mechanism.
Who coordinated?
What did they want?
What did each participant have to do?
Who knew the larger plan?
Why would those people continue cooperating?
What would happen if one defected?
And what should we observe if this mechanism actually existed?
Those questions lead directly into game theory.
What game theory can actually test
Game theory studies decisions in situations where one actor’s best choice depends partly on what other actors do.
The Stanford Encyclopedia of Philosophy describes game theory as the study of how interacting choices produce outcomes relative to the participants’ preferences or utilities. A Nash equilibrium is a set of strategies in which no player can improve their payoff by changing strategy alone while everyone else’s strategy remains fixed.
That gives us an important distinction.
If one person simply decides whether a benefit is worth a cost, we are mostly doing expected-utility analysis.
If Alice’s decision to remain silent depends on whether Bob will remain silent, Bob’s decision depends on whether Carol might cooperate with investigators, and investigators can change everyone’s incentives, we have a strategic game.
Conspiracies are therefore natural candidates for game-theoretic analysis because secrecy, cooperation, monitoring, betrayal and punishment are interdependent decisions.
But an equilibrium is not a prophecy.
Game-theoretic equilibria are analytical tools. Even the Stanford treatment cautions that equilibrium concepts need not describe what people will immediately do in the real world, especially when players have imperfect information, uncertain beliefs or limited rationality.
So the appropriate question is not:
“Did I find a Nash equilibrium, therefore this conspiracy happened?”
It is:
“Does the alleged mechanism contain a reasonably stable strategic configuration, and what would have to be true for that configuration to persist?”
The GAMES test: a framework for stress-testing a conspiracy hypothesis
The following GAMES test is a sherafy.com analytical framework synthesized from game theory, decision theory and the literature on conspiracy epistemology. It is not a published or clinically validated assessment instrument.
| Step | Question |
|---|---|
| G: Goal and alternatives | What exactly is the alleged conspiracy trying to accomplish, and what competing explanations could produce the same observations? |
| A: Actors and information | Who must participate, who must know the larger plan, who sees only fragments, and who could expose useful evidence? |
| M: Motives and payoffs | What does each critical actor gain or lose by cooperating, defecting, exposing the scheme or doing nothing? |
| E: Equilibrium and instability | Is continued cooperation actually a stable strategy, or does the alleged conspiracy require actors to behave against their own interests? |
| S: Signals and evidence | What observations should become more likely if the hypothesis is correct, and what observations should lower confidence in it? |
The value of the framework is not the acronym. It is the discipline it imposes.
A theory that sounds persuasive as a narrative can look very different after its assumptions are made explicit.
Start with the goal, not “who benefits?”
A common conspiracy argument begins with:
Who benefited?
That can be useful for generating hypotheses. It is nowhere near enough to establish causation.
The Internet Encyclopedia of Philosophy discusses this problem directly. A credible conspiracy explanation should identify plausible motives and should describe actions that would actually further the goals attributed to the conspirators.
Suppose Company A benefits enormously when Company B collapses.
That establishes a potential incentive.
It does not establish that Company A caused the collapse.
A stronger analysis asks whether secretly causing the collapse would produce a better expected outcome for Company A than its alternatives after accounting for cost, detection risk, legal liability, reputational damage and the possibility of failure.
The question changes from:
Who benefited?
to:
What strategy would a rational actor facing these incentives actually prefer?
That is considerably harder to answer, but far more useful.
Identify the people who actually need to know
Raw organization size is often a poor proxy for conspiracy size.
Imagine a company with 50,000 employees. A fraudulent decision secretly authorized by four executives does not become a 50,000-person conspiracy merely because thousands of employees perform ordinary work downstream.
For strategic analysis, the important population is closer to:
people who possess sufficiently incriminating knowledge or evidence to threaten the scheme.
That may include core decision-makers, operational participants, peripheral insiders, accountants, contractors, auditors or outside witnesses.
The distinction matters because secrecy depends on the distribution of information, not merely the number of people appearing on an organizational chart.
Compartmentalization can make a secret arrangement easier to protect because fewer participants understand the entire operation.
But compartmentalization has a cost: restricting information can also make complicated coordination harder.
That creates a genuine strategic tradeoff.
A conspiracy cannot simply assume both perfect secrecy and perfect coordination for free.
Build the payoff structure
For an important participant (i), define a simplified difference between remaining in the arrangement and defecting:
[ \Delta U_i = E[U_i(\text{cooperate})] – E[U_i(\text{defect})] ]
If (\Delta U_i) is strongly positive, continued cooperation may be individually attractive.
If it is strongly negative, the theory needs some explanation for why the participant remains cooperative.
The components could include:
[ E[U_i(\text{cooperate})]
B_i + F_i + L_i
D_i – P_i – C_i ]
where (B) could represent material benefits, (F) future benefits, (L) loyalty or status, (D) expected detection costs, (P) punishment if discovered and (C) other continuing costs.
Defection might instead include:
[ E[U_i(\text{defect})]
R_i + O_i + I_i
T_i – K_i – X_i ]
where (R) could represent a reward or leniency benefit, (O) outside opportunities, (I) personal incentives to expose the scheme, (T) retaliation risk, (K) career or reputational costs and (X) remaining legal exposure.
These variables should usually be treated as ranges or qualitative relationships, not invented numbers.
Writing “retaliation costs = $314,282” does not make an uncertain assumption scientific.
The useful question is sensitivity:
How much would this assumption have to change before the strategic conclusion changes?
If the entire conspiracy becomes unstable when one uncertain variable moves slightly, the hypothesis is fragile.
If it remains stable across a broad range of reasonable assumptions, that tells us something different.
A simple example: would someone fake their location for an obscure blog?
Consider a deliberately boring example.
An obscure website wants an approximate picture of where its visitors live. It uses IP-derived location data.
Someone correctly points out:
“A visitor could use a VPN, so the location might be wrong.”
That establishes a technical possibility.
It does not yet establish deliberate manipulation.
Some visitors will already use VPNs for privacy, work or convenience. Their apparent locations may therefore be wrong even though they have no interest in fooling the website.
That is ordinary measurement error.
Now ask the strategic question:
What does a visitor gain by intentionally making an obscure website think they are in Phoenix instead of Long Beach?
If the answer is essentially nothing, purposeful manipulation has little payoff for an ordinary visitor.
It remains possible.
But capability and incentive are different variables.
Now change one thing
Suppose the website announces:
California respondents receive $500.
The technology has not changed.
VPNs have not changed.
The ability to spoof location has not changed.
The payoff changed.
Now some visitors have a meaningful reason to appear Californian.
The site’s incentives change too. Spending resources on location verification may suddenly become rational because manipulation now produces a measurable cost.
The strategic interaction becomes:
| Visitor | Website | Likely consequence |
|---|---|---|
| Little reward for spoofing | Verification is costly | Minimal adversarial verification |
| High reward for spoofing | No verification | Greater manipulation incentive |
| High reward for spoofing | Strong verification | Spoofing becomes less attractive |
| Penalty for detected fraud | Strong verification | Manipulation can become strongly unattractive |
This simple example illustrates one of the most useful lessons game theory brings to conspiracy analysis:
The existence of a capability does not tell us how frequently actors will deliberately use it. Incentives help determine that.
Real conspiracies are already fought by changing the payoff for betrayal
Game theory is not being added to conspiracy analysis merely because it sounds mathematical.
Authorities already use strategic incentives to destabilize documented conspiracies.
The U.S. Justice Department’s Antitrust Division Leniency Policy provides non-prosecution protections under specified conditions to qualifying organizations and individuals that disclose participation in criminal antitrust conspiracies and cooperate with investigators. The department explicitly describes predictable incentives for voluntary disclosure as a tool that has helped uncover domestic and international cartels.
The policy changes the game.
Before leniency, cartel member A may compare:
remain silent and preserve cartel profits
with
confess and face prosecution anyway.
After a sufficiently valuable leniency option appears, the second strategy changes:
confess early, cooperate and potentially receive protection while competitors face investigation.
Now every conspirator has to worry that another conspirator may defect first.
Nathan Miller examined this mechanism in a 2009 American Economic Review paper, developing a theoretical model of cartel behavior and comparing it with two decades of enforcement data. His statistical tests were consistent with leniency improving cartel detection and deterrence.
This gives us a real-world lesson:
Secrecy is not simply a personality trait possessed by conspirators. It can be an equilibrium maintained or destroyed by incentives.
That is why “why hasn’t somebody talked?” is too weak a question.
The better question is:
What happens to a knowledgeable participant who talks, and how does that payoff compare with remaining silent?
“Someone would have talked” is useful, but too simplistic
A famous mathematical attempt to analyze conspiracy secrecy came from physicist David Robert Grimes in 2016.
His PLOS ONE model estimated how conspiracy failure risk changes with variables including the number of conspirators and time. Under its assumptions, large long-running conspiracies become increasingly difficult to maintain.
But the paper contains a limitation particularly relevant to the framework proposed here.
Grimes explicitly states that the model does not account for the dynamics, motivations and interactions of individual agents and suggests agent-based modeling as a possible way to incorporate internal pressures and differing actor characteristics.
That limitation matters.
Two conspiracies containing the same number of people can have radically different stability.
One may involve highly compensated participants with shared legal exposure, powerful monitoring and severe retaliation risks.
Another may involve poorly rewarded participants who possess copies of incriminating records, dislike one another and are offered immunity for cooperating.
A headcount alone cannot capture that difference.
There are also philosophical criticisms of relying too heavily on the “loose lips” argument. Ryan Ross argued in Episteme that arguments based on secrecy have difficulty ruling out small conspiracies and cannot substitute for directly examining the alleged positive evidence offered for a specific claim.
So the better variable is not simply:
How many people were involved?
It is:
How many people possessed sufficiently meaningful knowledge or evidence, for how long, under what incentives, with what opportunities to expose it?
That is a much richer question.
Why conspirators sometimes do keep cooperating
Another mistake would be assuming game theory predicts automatic betrayal.
It does not.
Repeated games can support cooperation that would collapse in a one-time interaction. The Stanford Encyclopedia’s treatment of game theory notes that repeated versions of even the Prisoner’s Dilemma can support cooperative Nash equilibria that do not exist in the simple one-shot version.
Real secret arrangements can also create mechanisms that make cooperation attractive.
Participants may expect future financial benefits. They may share legal liability. They may fear retaliation. They may value loyalty or status. They may depend professionally on the same organization. Participants can monitor one another. Exposure may destroy the whistleblower’s own career or implicate the whistleblower in the underlying misconduct.
None of those mechanisms proves a conspiracy exists.
They simply show why:
“Someone would obviously talk”
is not a complete argument.
Sometimes defection is rational.
Sometimes silence is.
The model has to establish which situation it is actually describing.
A conspiracy hypothesis should compete against real alternatives
Testing only the conspiracy hypothesis creates another problem: almost any story can appear persuasive if its only competitor is “nothing unusual happened.”
The analysis should instead compare multiple causal explanations.
Suppose an organization behaves strangely.
One hypothesis is secret coordination.
Another is that employees independently face similar incentives.
That distinction matters enormously.
Imagine five companies raising prices shortly after the same input cost rises.
Parallel behavior exists.
But coordination has not yet been established. Five decision-makers responding independently to the same economic incentive can produce superficially coordinated behavior.
Now suppose investigators discover private communications in which executives agree on prices and establish punishment for anyone who defects.
The coordination hypothesis predicts that evidence much more naturally.
The important question therefore becomes:
Which hypothesis makes the observed evidence less surprising?
That is where game theory begins to connect with Bayesian reasoning.
Use Bayesian updating after the strategic model
Game theory asks:
If these actors had these incentives and information, what behavior should we expect?
Bayesian reasoning asks:
Now that we observed something, should it make us more or less confident in one explanation relative to another?
A useful conceptual form is:
[ P(H\mid E) \propto P(E\mid H)P(H) ]
You do not need to invent exact percentages to use the logic.
The useful comparison is often simply:
Would this evidence be substantially more expected if H1 were true than if H2 were true?
A 2026 PNAS Nexus study is relevant here for a different reason. Researchers examining conspiracy beliefs found a relationship between conspiracism and assigning incoherent levels of belief to incompatible explanations. The authors argue for treating beliefs probabilistically rather than merely as binary “believe/disbelieve” judgments.
That is a useful discipline for anyone investigating controversial claims.
You can reasonably entertain several hypotheses simultaneously.
But confidence assigned to mutually incompatible explanations must remain coherent.
If evidence weakens one hypothesis, another should generally become relatively more important. We should not simply add each new contradictory theory to an ever-growing collection of things we are simultaneously certain happened.
The hardest problem: theories that explain away missing evidence
Secret coordination creates a genuine epistemic complication.
If evidence exists, a believer can say:
“There is the evidence.”
If evidence does not exist, a sufficiently flexible theory can say:
“Of course there is no evidence. The conspirators destroyed it.”
That does not make the second statement logically impossible.
Real conspirators sometimes do destroy evidence.
The problem arises when hidden mechanisms are repeatedly added specifically to protect a theory from observations that otherwise count against it.
A September 2026 Synthese paper by James H. McIntyre proposes a hidden evidence penalty. The argument, presented as a philosophical position rather than an established empirical law, is that explanations incur increasing epistemic costs as they rely on increasingly undetectable structures. McIntyre explicitly frames the position as a way to preserve case-specific investigation without treating hiddenness as costless.
The insight can be expressed simply:
“They successfully hid all evidence” is an additional assumption. It does not get added to a theory for free.
Consider two hypotheses.
H1 predicts that if it were true, documents, witnesses or financial traces would probably exist.
After an extensive investigation, none are found.
H1 can survive logically by adding:
Every relevant record was destroyed, every witness remained silent and every investigative path was successfully obstructed.
Possible?
Perhaps.
But H1 is now carrying several additional assumptions.
That should normally reduce its attractiveness relative to a competing hypothesis that predicted the absence of those traces from the beginning.
This is not because every true theory must be easy to prove.
It is because failed predictions should cost something.
What should count against a conspiracy hypothesis?
A useful framework must allow the answer to move in both directions.
Confidence in a conspiracy hypothesis should generally fall when actors required for the theory have strong incentives to defect and no credible mechanism prevents it; when the theory repeatedly expands the number of secret participants to explain failed predictions; when alleged motives do not actually make the actors better off; when supposedly necessary coordination leaves none of the traces the mechanism predicts; when the theory relies increasingly on undetectable mechanisms; or when a competing explanation accounts for the same observations with fewer unsupported assumptions.
Those are not automatic disproofs.
They are evidentiary costs.
The opposite evidence can matter too.
Independent documentary corroboration, communication records, financial transfers, credible cooperating insiders, evidence of monitoring or retaliation, admissions, coordinated actions difficult to explain independently, or records establishing the alleged mechanism can strengthen a conspiracy hypothesis.
The important point is symmetry of method:
evidence must be allowed to move confidence down as well as up.
The conventional explanation has to survive the same test
The framework should not become an elaborate machine for dismissing unconventional claims.
The conventional explanation gets no exemption.
If an official or widely accepted explanation requires several actors to repeatedly behave against their incentives, contradicts established chronology, fails to predict important evidence or survives only through a growing collection of ad hoc explanations, those weaknesses matter too.
The questions remain the same:
Does the mechanism make sense?
Are the required actors identified?
Do their incentives fit?
What observations does the explanation predict?
Which observations contradict it?
What assumptions are necessary to rescue it?
How does it perform against competing explanations?
That is a much stronger standard than treating either institutional authority or outsider suspicion as evidence in itself.
What game theory cannot tell you
Game theory becomes dangerous when its mathematical appearance is mistaken for knowledge.
We rarely know someone’s true utility function.
Human beings do not always maximize expected value.
Fear, ideology, friendship, spite, morality, incompetence and misunderstanding can all affect behavior.
Actors may possess different information.
They may misjudge one another.
Some participants may not understand the larger game at all.
And researchers can assign whatever numbers they want to a model.
If those inputs are invented, the resulting decimal places are decoration.
A “conspiracy probability calculator” announcing that a theory is 73.6% likely to be true would therefore be misleading unless those probabilities came from independently defensible empirical estimates and a validated model.
The more responsible use of simulation is sensitivity analysis:
Under what assumptions is the alleged arrangement stable, and which assumption causes the conclusion to reverse?
That is useful even when the inputs remain uncertain.
For example:
Stable under almost every plausible payoff assumption tells us something.
Stable only if every participant accepts enormous personal risk for almost no benefit tells us something else.
Neither result establishes historical truth.
Evidence still has to do that.
A better standard for conspiracy claims: calibrated skepticism
The most useful outcome of this framework is not greater belief in conspiracies.
It is better discrimination.
A conspiracy hypothesis should not receive special credibility because it explains something mysterious.
Nor should it receive automatic rejection because secret coordination sounds uncomfortable or because “conspiracy theory” has become a dismissive label.
A strong hypothesis should specify a coherent mechanism, survive a serious analysis of actor incentives, produce observations that differ from its competitors, explain existing evidence better than reasonable alternatives and remain vulnerable to evidence that could lower our confidence.
That standard is demanding.
It should be.
The 2026 research on conspiracy discrimination provides an unusually appropriate closing lesson: better reasoning was associated not merely with saying “no” to conspiracies, but with doing a better job distinguishing documented conspiracies from unsupported ones.
Game theory gives us one additional tool for doing that.
It asks something ordinary discussions too often skip:
If this secret arrangement really existed, why would the people inside it behave the way the theory requires?
Sometimes the answer will expose a major weakness.
Sometimes it will reveal a surprisingly stable mechanism.
Either way, that is more informative than beginning with either belief or ridicule.
References and Further Reading
Conspiracy Theories, Definitions and Evidence
Conspiracy Theories — Internet Encyclopedia of Philosophy
A detailed overview of the philosophical debate over conspiracy theories, including definitions, motives, evidence, scale and standards for evaluation.
Defining “Conspiracy Theory” Without Begging the Question: A Roadmap — Erkenntnis
Romy Jaster and Geert Keil’s 2026 examination of generalist and particularist approaches to conspiracy theories. Especially useful for understanding why case-by-case investigation can coexist with general epistemic caution.
Rejecting Conspiracies Doesn’t Necessarily Mean That You’re Smart: The Relationship Between Conspiracy Discrimination and Reasoning Ability — British Journal of Social Psychology
A 2026 three-study investigation distinguishing the ability to discriminate between verified and unverified conspiracies from a general tendency to believe or reject conspiracy claims.
Conspiracy Theories and the Hidden Evidence Penalty — Synthese
James H. McIntyre’s 2026 argument that explanations relying increasingly on undetectable structures incur an epistemic cost. Useful for understanding why hidden evidence cannot be invoked without consequence.
Still Very Much Dead and Alive: Incoherence in Conspiracism as a Departure From Bayesian Rationality — PNAS Nexus
A 2026 probabilistic analysis of belief in mutually incompatible conspiracy explanations and Bayesian coherence.
Game Theory and Strategic Incentives
Game Theory — Stanford Encyclopedia of Philosophy
A comprehensive introduction to game theory, strategic interaction, Nash equilibrium, repeated games, information and the limitations of equilibrium as a predictive concept.
Antitrust Division Leniency Policy — U.S. Department of Justice
The Justice Department’s current framework for incentivizing qualifying cartel participants to self-report and cooperate, providing a real-world example of changing the payoff for defection within a documented conspiracy.
Strategic Leniency and Cartel Enforcement — American Economic Review
Nathan H. Miller’s game-theoretic and empirical study of cartel leniency, finding results consistent with enhanced detection and deterrence.
Conspiracy Size, Secrecy and Defection
On the Viability of Conspiratorial Beliefs — PLOS ONE
David Robert Grimes’ 2016 mathematical model of conspiracy failure over time. Particularly relevant because the paper itself notes that its model does not incorporate individual motivations and interactions and suggests agent-based modeling as a future extension.
Can You Keep a Secret? BS Conspiracy Theories and the Argument From Loose Lips — Episteme
Ryan Ross’ analysis of arguments that large conspiracies should inevitably leak, including limitations involving small conspiracies and the need to examine alleged positive evidence rather than secrecy alone.
Editorial currency note: Academic literature on conspiracy epistemology and psychology is developing rapidly. The research and government policy sources above were reviewed for this article in September 2026. Government enforcement policies and newer research findings may change.
