By Augustus Kane · Published May 13, 2026 · Updated May 13, 2026
A retired Federal Bureau of Investigation agent reads the same ten names a Reddit thread reads and sees five of them as a pattern. The cognitive psychologists who have spent fifty years on this problem read those ten names and ask a different question: how many deaths per year would we expect among American specialists working in nuclear, aerospace, and biodefense research, and is ten across a multi-year window unusually large or unusually small against that base rate? The argument that follows is not that the public is wrong to ask the question. It is that the question, asked properly, almost always answers itself in a direction the cluster-frame cannot accommodate [1][2].
Direct Answer: Why the missing-scientists pattern fails statistical scrutiny
The April 2026 FBI inquiry into roughly ten deaths and disappearances among American scientists looks like a coordinated pattern to readers running the representativeness heuristic, but it fails three statistical tests: base-rate neglect (deaths in this population are expected, not exceptional), selection bias (the ten cases were chosen because they fit, not collected at random), and the absence of attribution evidence that characterized real coordinated campaigns such as the 2007 to 2012 Iranian nuclear-scientist assassinations [1][2][3].
The CNN reporting and the cluster as the public first met it
On April 21, 2026, CNN reported that at least ten individuals tied to sensitive United States research had died or disappeared in recent years, prompting a federal investigation led by the Federal Bureau of Investigation in coordination with the Department of Energy and the Department of War. The roster included researchers connected to nuclear laboratories, aerospace programs, and adjacent classified work. Within twenty-four hours the cluster had migrated from CNN’s homepage to social-media threads, podcast monologues, and a House Oversight Committee announcement that the body would open its own inquiry [1].
The shape of the story is familiar to anyone who reads the historical record on conspiratorial clusters. A federal agency opens an investigation. The opening of the investigation is treated, in public discussion, as a substantive finding of suspicious linkage. The two are not the same act. An FBI inquiry that finds nothing also exists, has historically been common, and is in fact the modal outcome when a federal agency probes a cluster that an outside observer has assembled by selection.
Why pattern-recognition fires first and asks questions later
Amos Tversky and Daniel Kahneman, in the foundational papers collected in Judgment Under Uncertainty (Cambridge, 1982), described a family of cognitive shortcuts the human mind uses when probability is hard and intuition is fast. Two of those shortcuts matter here. The representativeness heuristic asks whether a candidate explanation resembles the kind of thing that would produce the observed event; if it does, the mind tends to assign it a probability higher than the underlying data support. The base-rate fallacy describes the systematic neglect of the prior probability of the explanation, in favor of attention to surface features of the case [2].
Applied to the missing-scientists frame, the heuristic runs as follows. Coordinated state-actor assassination of scientists is a thing that resembles what would produce a cluster of dead scientists with sensitive clearances. Therefore the mind, running fast, assigns the explanation a probability disproportionate to its actual frequency in the universe of explanations for why specialists die or disappear. The base rate of coordinated assassination among American researchers is, on the historical record, vanishingly small. The base rate of unrelated death, suicide, accident, and personal disappearance among the same population over a multi-year window is not small. The intuition runs faster than the arithmetic, and the arithmetic is the part that ought to control.
The salience asymmetry the news cycle introduces
A second mechanism compounds the first. Deaths that fit the conspiratorial frame are remembered, recirculated, and aggregated by interested observers. Deaths that do not fit the frame are not collected at all. The ten names in the April 2026 reporting are not a random sample from the population of American scientists who have died in recent years; they are the names someone selected because they appeared selectable. This is the classical structure of a selection-biased sample, and any inference drawn from it about the underlying population is, in the technical sense, uninterpretable.
The arithmetic of “ten deaths in a population of N specialists”
The exact denominator is contested because the relevant population is differently defined by different observers, but the order of magnitude is not. Department of Energy national laboratories alone employ roughly forty thousand scientists, engineers, and technical staff. Adding NASA, the major defense contractors, and adjacent classified-program personnel produces a working figure on the order of one hundred thousand individuals over the multi-year window the cluster covers. American adult mortality rates, even restricted to working-age professionals, run on the order of two to three deaths per thousand per year. The expected number of deaths in this population, across the three-to-four-year window the April 2026 cluster spans, is therefore in the hundreds, not in single digits [3].
Ten deaths is not the anomaly. Ten deaths is what we would expect to find at the lower bound of the random fluctuation around the expected number. The anomaly, if there is one, would be a sustained excess above that base rate, demonstrated by a complete enumeration of all deaths in the relevant population over the relevant window, not by selecting the deaths that fit a narrative shape and ignoring the rest.
Medical sociologist Robert Bartholomew, quoted in Scientific American’s coverage, named the cognitive mechanism by its technical term: apophenia, the tendency to perceive meaningful patterns in unrelated events. The point is not that the readers running the heuristic are foolish. The point is that the heuristic is misfiring on a class of input for which it was not evolved [2].
The microbiologist cluster of 2001 to 2002: a closely studied precedent
The argument is not hypothetical. The historical record contains at least one prior cluster of this exact shape, and it has been studied retrospectively long enough to assess what it actually was. In the fourteen months following the September 11 attacks and the anthrax letters of October 2001, roughly twenty microbiologists and bio-weapons researchers died in the United States, the United Kingdom, Australia, and Russia. The names entered the conspiratorial canon: Dr. Benito Que, Dr. Don Wiley, Dr. Vladimir Pasechnik, Set Van Nguyen, Dr. Ian Langford, Dr. Tanya Holzmayer, and others. The most politically significant case in the cluster was that of Dr. David Kelly, the British weapons inspector whose death in July 2003 became the subject of the Hutton Inquiry [4].
Retrospective analysis of the microbiologist cluster, conducted by epidemiologists, journalists, and the Hutton Inquiry itself in Kelly’s specific case, produced a consistent reading. The cluster was a selection artifact. Microbiology is a populous field; the global community of working microbiologists numbers in the tens of thousands; the expected number of deaths over a fourteen-month window across that population is consistent with the observed count; and the individual causes of death, examined one at a time, included car accidents, falls, heart attacks, suicides, and a small number of unsolved cases of the sort any large professional population accumulates. Lord Hutton’s January 2004 report concluded that Kelly’s death was a suicide. The conclusion has been contested by some commentators and reaffirmed by subsequent reviews; the contest itself illustrates the durability of the conspiratorial frame even when the closest examination available finds nothing to support it [4].
The Iranian nuclear-scientist assassinations: what a real coordinated pattern looks like
For the comparison to be honest, the contrasting case must be put on the table. Between 2007 and 2012, at least five Iranian nuclear scientists were killed in attacks widely attributed to Israeli intelligence services. Masoud Alimohammadi was killed by a motorcycle bomb in Tehran in January 2010. Majid Shahriari was killed by a magnetic device attached to his car in November 2010. Darioush Rezaeinejad was shot in July 2011. Mostafa Ahmadi Roshan was killed by a magnetic bomb in January 2012. Mohsen Fakhrizadeh, the senior figure in the program, was killed in November 2020 by a remote-operated weapon, on what the New York Times subsequently reported as Mossad’s confirmed account [5].
Three features of that campaign distinguish it from the April 2026 American cluster, and they are the features the cognitive-bias literature predicts a reader should look for before assigning the coordinated-pattern frame. First, modus operandi consistency: the Iranian assassinations used a recurring tactical signature (motorcycle-mounted explosives, magnetic devices, daylight attacks in Tehran) that connected the cases to one another beyond the bare fact of the victims’ profession. Second, attribution evidence: Iranian intelligence services made specific public accusations, named operatives, broadcast confessions, and produced documentary evidence; Western intelligence services and U.S. officials separately confirmed Israeli involvement on background; the Times of Israel and the New York Times later published sourced accounts. Third, plausible motive at the state level: a sustained Israeli policy of preventing Iranian nuclear progress, openly debated in the Knesset and the security cabinet, provided a documented institutional context for the campaign [5].
The April 2026 American cluster has none of these features. The methods of death, where determined, are heterogeneous (illness, suicide, accident, missing-person). No state actor has claimed responsibility or had responsibility imputed by a competing intelligence service. No documentary chain of command has surfaced. The cases are connected to one another by the bare fact of the victims’ professional sector, which, in a population this large, is what selection bias guarantees one will find.
FBI investigating is not FBI confirming
A federal investigation that finds nothing also exists. This is a sentence that the public discussion of the missing-scientists story has, on the evidence, struggled to absorb. The FBI’s spearheading of the inquiry, announced in April 2026, is a procedural act in response to public pressure and congressional interest. It is not a substantive finding that the cluster is real, that the cases are connected, or that foul play is more likely than not. CBS News, in its own reporting, found that none of the energy-security and law-enforcement experts it interviewed saw an obvious link between the cases. CNN’s own sourced experts rejected the coordinated-pattern interpretation. The FBI itself, in its public statements, has described the inquiry as a search for connections, not a confirmation of them [1][6].
Joseph Uscinski, in American Conspiracy Theories (Oxford, 2014), documented the consistent finding that the opening of an official inquiry is interpreted by motivated readers as validation of the conspiratorial hypothesis the inquiry was opened to test. The interpretation runs backward through the procedural logic. Inquiries are opened when claims are made; the opening does not adjudicate the claim; the closing, if it comes, will. The cognitive shortcut that treats the opening as the closing is documented in the literature; it is not a personal failing of any individual reader [7].
The cognitive-bias bibliography for the curious reader
The relevant scholarly literature is not obscure and not new. Amos Tversky and Daniel Kahneman’s 1974 paper, “Judgment Under Uncertainty: Heuristics and Biases,” reprinted in their 1982 Cambridge volume of the same name, is the canonical source on the representativeness heuristic and the base-rate fallacy. Cass Sunstein and Adrian Vermeule’s 2008 paper, “Conspiracy Theories,” and Sunstein’s 2014 book Conspiracy Theories and Other Dangerous Ideas (Simon & Schuster), address the propagation mechanisms once a misfiring heuristic has produced an initial belief. Joseph Uscinski and Joseph Parent’s American Conspiracy Theories (Oxford, 2014) and Uscinski’s edited 2018 volume Conspiracy Theories and the People Who Believe Them (Oxford) provide the empirical political-science framing. Anna Merlan’s Republic of Lies (Metropolitan, 2019) and Jonathan Kay’s Among the Truthers (Harper, 2011) are the journalistic accounts that locate the same patterns in the modern American media ecosystem [2][7][8].
The literature converges on a finding that deserves to be stated plainly. Conspiratorial pattern-recognition is not a sign of stupidity; it is a sign that a fast cognitive system trained on small-scale social environments is being asked to evaluate a large-scale statistical environment. The mismatch is structural. The remedy is the slow work of asking what the base rate is, how the sample was drawn, what the attribution evidence shows, and what the historical analogues actually produced on retrospective examination.
A reading discipline, not a dismissal
Nothing in the argument above forecloses the possibility that one or more of the cases in the April 2026 cluster, examined individually, will turn out to involve foul play. The proposition the argument forecloses is the stronger one: that the cluster as a whole demonstrates a coordinated pattern. The individual investigations will produce individual findings, and the individual findings, when they come, will deserve the same scrutiny the cluster has been offered here. For the case-by-case evidence ledger that grounds this framing analysis, see Iris Kowalczyk’s companion piece on the same investigation. For the wider catalogue of conspiracy patterns this site tracks, the Conspiracy Theories and Secret Societies pillar is the entry point. The conspiracy, if it is a conspiracy, will be in the document. It will not be in the headline.
Frequently asked questions
Is the FBI’s April 2026 inquiry a confirmation that the missing-scientists pattern is real?
No. An FBI inquiry is a procedural search for connections, not a substantive finding that connections exist. CNN’s own sourced experts rejected the coordinated-pattern interpretation, and CBS News reported that energy-security and law-enforcement specialists it interviewed saw no obvious link between the cases. Federal inquiries that find nothing are historically the modal outcome for clusters of this shape [1][6].
What is the representativeness heuristic and why does it matter here?
It is a cognitive shortcut described by Amos Tversky and Daniel Kahneman in 1974, in which probability is judged by the resemblance of a candidate explanation to the kind of thing that would produce the observed event. Applied to the missing-scientists frame, it leads readers to assign coordinated assassination a probability disproportionate to its actual frequency in the universe of explanations [2].
What is base-rate neglect?
The systematic failure to take into account the prior probability of an explanation when evaluating a specific case. In the missing-scientists context, it shows up as readers asking whether ten deaths look like a lot without asking how many deaths in the relevant population would be expected at baseline. The expected number, across the relevant scientific workforce and time window, runs into the hundreds [2][3].
How does the 2001 to 2002 microbiologist cluster compare?
It is the closest historical analogue. Roughly twenty microbiologists died across a fourteen-month window following 9/11; the cases were assembled into a conspiratorial cluster; retrospective analysis, including the Hutton Inquiry into David Kelly’s specific death, attributed the cluster to selection bias and base-rate effects rather than coordinated foul play [4].
How does the Iranian nuclear-scientist case differ?
It is the contrasting case in which the coordinated-pattern frame was supported by evidence. The 2007 to 2012 assassinations of at least five Iranian nuclear scientists showed modus operandi consistency, attribution by competing intelligence services, and a documented state-level motive. The April 2026 American cluster has none of those features [5].
Who are the academic authorities on this kind of pattern-recognition error?
Amos Tversky and Daniel Kahneman on heuristics and biases; Cass Sunstein on conspiracy propagation; Joseph Uscinski and Joseph Parent on the empirical politics of American conspiracy theories; Anna Merlan and Jonathan Kay on the journalistic mapping of contemporary conspiracy ecosystems [2][7][8].
Does this argument dismiss the families of the deceased?
No. Individual cases of unexplained death deserve individual investigation, and the families’ interest in answers is legitimate and unrelated to the cluster-level inference. The argument here is that the cluster-as-coordinated-pattern frame is unsupported by the available evidence, not that the individual cases are unworthy of attention.
What would change the assessment?
Three things, drawn from the Iranian comparison. Modus operandi consistency across the cases. Attribution evidence, whether by claim, capture, or third-party intelligence reporting. Documentary evidence of state-level or organizational motive. None of those is currently in the public record for the April 2026 cluster. If any of them surfaces, the inference shifts.
What is apophenia?
A technical term for the perception of meaningful patterns in unrelated events. The medical sociologist Robert Bartholomew, quoted by Scientific American on the April 2026 cluster, used the term to name the cognitive mechanism by which random or selection-biased data acquires the appearance of design [2].
Sources
[1] CNN, “At least 10 people tied to sensitive US research have died or disappeared in recent years, sparking federal investigation,” April 21, 2026, cnn.com.
[2] Amos Tversky and Daniel Kahneman, “Judgment Under Uncertainty: Heuristics and Biases,” Science 185 (1974): 1124-1131, reprinted in Kahneman, Slovic, and Tversky, eds., Judgment Under Uncertainty: Heuristics and Biases (Cambridge University Press, 1982); Robert Bartholomew quoted in Scientific American, “Math and statistics help explain the FBI’s missing scientists cases,” April 2026.
[3] National Center for Health Statistics, “Mortality in the United States,” CDC NCHS Data Brief series, 2020-2025; Department of Energy, “National Laboratory Workforce Data,” DOE Office of Science annual reports.
[4] Lord Brian Hutton, Report of the Inquiry into the Circumstances Surrounding the Death of Dr. David Kelly C.M.G. (HC 247, January 28, 2004); Wikipedia, “David Kelly (weapons expert),” retrieved May 2026; contemporary reporting in The Guardian, The Times (London), and the BBC, 2001-2004.
[5] “Assassinations of Iranian nuclear scientists,” Wikipedia, retrieved May 2026; New York Times, “Israel’s Killing of Iran Atomic Scientist Was Years in the Making,” September 18, 2021; Times of Israel, “Mossad killed Iran’s top nuke scientist with remote-operated machine gun — NYT,” September 19, 2021.
[6] CBS News, “FBI investigating deaths and disappearances of staff at secretive government laboratories. Here’s what we know,” April 2026.
[7] Joseph E. Uscinski and Joseph M. Parent, American Conspiracy Theories (Oxford University Press, 2014); Joseph E. Uscinski, ed., Conspiracy Theories and the People Who Believe Them (Oxford University Press, 2018); Cass R. Sunstein and Adrian Vermeule, “Conspiracy Theories,” Journal of Political Philosophy 17, no. 2 (2009): 202-227.
[8] Anna Merlan, Republic of Lies: American Conspiracy Theorists and Their Surprising Rise to Power (Metropolitan Books, 2019); Jonathan Kay, Among the Truthers: A Journey Through America’s Growing Conspiracist Underground (Harper, 2011); Cass R. Sunstein, Conspiracy Theories and Other Dangerous Ideas (Simon & Schuster, 2014).
