Abstract
The question this article answers: why does the most calculating person at work so rarely leave a trace, and why is the research on them so much weaker than the research on every other dark trait?
Machiavellianism is the least well-measured and most confidently discussed member of the dark triad, the standard grouping of Machiavellianism, narcissism, and psychopathy, and of the dark tetrad, the same grouping with everyday sadism added. This article appraises what the trait predicts in everyday working behaviour, and it treats the measurement problem as the subject rather than as an apparatus note, because on the evidence assembled here the two are the same problem.
It opens with the strongest available separation of Machiavellianism from psychopathy, a thirty-day study of 317 adults using experience sampling, meaning a method that catches conduct close to when it happens rather than asking for a summary months afterwards, in which the two constructs overlapped by more than seventy per cent when assessed as traits and by roughly sixteen per cent when tracked as daily states, with manipulative days predicting subsequent impulsive days in one direction only (Walczak, Rogoza & Jones, 2026). It then examines the findings this article was originally built on. A survey of 812 students found narcissism, psychopathy and sadism associated with generative AI misconduct while Machiavellianism was not (Sun, Tang, Zhou, Loan & Wang, 2025), and the article reports the authors’ own account of that empty finding, which rests on under-reporting and on a failure of their Machiavellianism items, sets it against a study of 283 people reaching the opposite conclusion (Greitemeyer & Kastenmüller, 2023), and locates both inside a narrative review of twenty-three studies (Jones & Jones, 2026). All three rely on self-report, a method that asks someone to describe their own conduct and that this territory is least well served by.
The measurement argument is developed through two records. Three studies concluded that Machiavellianism and psychopathy assessments produce nearly identical profiles and that the former fails to capture the construct its own theory describes (Miller, Hyatt, Maples-Keller, Carter & Lynam, 2017). A subsequent instrument-development programme across 4,607 participants reduced an overlap that two syntheses had placed at .58 and .52 to between .23 and .25 (Grosz, Harms, Dufner, Kraft & Wetzel, 2020).
Workplace consequences are traced through two multi-source studies from the same research group: a two-study programme of 1,438 participants in which political skill masked the trait in colleague ratings while the harmful conduct continued (Blickle et al., 2020), and a study of 550 targets rated by 1,127 coworkers in which reputation carried the effect and political skill removed it (Kückelhaus, Meurs & Blickle, 2024). Occupational attainment is examined through 8,587 records from a German panel study in which the trait predicted holding a management position, monotonically and with the causal arrow running from personality to attainment (Baktash & Jirjahn, 2026). Aggregate workplace conduct is taken from a review of twenty-one papers (Duradoni et al., 2025).
A dedicated late chapter sets out the five kinds of people most often wrongly labelled with this trait, which in this territory includes the politically skilled colleague who is doing nothing of the kind. The practical section offers four moves and three refusals. The conclusion is that this is the one dark trait where the honest advice is not about handling a person at all, but about what an organisation counts as evidence.
Keywords: Machiavellianism; dark triad; dark tetrad; experience sampling; measurement overlap; political skill; workplace conduct; organisational psychology.
PART I. WHAT ACTUALLY HAPPENS
Chapter 1: The Colleague With Nothing on File
1.1 Three Rooms
The scenarios below are composite. They draw on recurring patterns in research and advisory practice, and they describe no real, identifiable person or organisation.
A finance director has been in post for six years and has never had a complaint made against her, which the human resources team will tell you without being asked, because they have had to look it up more than once, and the three people who have held the role reporting into her all left within eighteen months, each for a reason that made sense at the time and none of which was her. What her peers say, when they say anything at all, is that she is very good in the room and that it is difficult to describe afterwards what was agreed.
A programme lead at a manufacturer is the person everybody wants on the steering group, because he is unfailingly reasonable, never raises his voice, and has a gift for finding the version of a proposal that everybody can live with. Two of the four workstream owners have separately worked out that the version everybody can live with tends to be the one that leaves his own scope untouched, and neither of them has said so, because the observation sounds petty the moment you try to put it into words and because he has never actually done anything.
A partner in a professional services firm has a reputation for generosity with junior staff that is entirely deserved and entirely visible, and a habit, noticed only by the two other partners who have run large accounts alongside him, of being unavailable at precisely the moments when a decision would have had to carry his name, and neither of them has raised it. Both have independently started copying a third partner into anything that might later be contested.
1.2 What the Three Have in Common
Three different organisations, three different levels of seniority, and the same shape running through all of them: somebody has arrived at a firm conclusion about a colleague and cannot point to a single thing that would survive being said out loud. I started with three rather than one because the shape is not a property of a particular sector or grade, and because what they share is easy to misread as a failure of the people doing the observing. It is not. In each case, the observation is probably correct, the observers are experienced, and the reason they have nothing to show is not that they have been careless about gathering it.
They share a fourth thing, and it is the reason this article exists rather than being a chapter of the last one. In every case the file is clean, and in every case the absence of a file has already been used by somebody as an argument.
1.3 Why Reaching for the Label Fails Faster Here Than Anywhere Else
Every article in this series warns against reaching for a diagnostic label, and I have made that argument twice already about narcissism, where the cost of the label is that it starts a dispute about your judgement rather than a conversation about somebody’s behaviour. Here the cost is different, and I would say it is considerably steeper.
With the loud traits, the label is unnecessary because the behaviour is available: you can describe what happened in a meeting and the description will do the work the label was going to do. With this one, the label is the only thing you have, which is exactly why the temptation to use it is strongest and exactly why using it is worst. You will be asked what you are basing it on; you will have nothing, and the nothing will then be entered into the record as a finding about you.
A second cost lands on other people. Machiavellian has drifted in ordinary speech until it means roughly good at office politics, and Chapter 7 is about the four or five kinds of colleague who get caught by that drift, most of whom are simply competent at reading a room and have no unusual interest in anybody’s downfall. I have put that chapter late deliberately, because a caution absorbs better once a reader has some reason to care what it is protecting.
What the research offers you instead is not a better label and not a detection method. It offers three things, and I think the third is the one that will change what you do: a mechanism, which is restraint rather than deception; a measurement problem, which turns out to be the same problem your organisation has; and a route, meaning the specific and rather narrow channel through which this trait has ever been shown to cost anybody anything at work.
None of that requires you to conclude anything about a person, and none of it could. It requires you to notice a pattern across time, which is the one instrument this trait is not well defended against.
Chapter 2: Thirty Evenings
2.1 The Question Nobody Could Answer From a Questionnaire
For about a decade, the most serious argument in this literature has been whether Machiavellianism exists as a separate thing at all, and I want to state that plainly at the outset rather than hedge, because it is not a technical quibble and it governs how much weight anything else in this article can bear. The theoretical description is clear enough and has been for decades. A Machiavellian person is strategic, patient, willing to use others instrumentally, and above all controlled: they weigh the cost of being caught, and they act accordingly. A psychopathic person, in theory, wants similar things and does not weigh anything.
The trouble is that when researchers built questionnaires to measure the first, the questionnaires behaved like the second. Chapter 4 sets out how badly. For now, the useful thing to know is that this left the field with an obvious question it had no clear way to answer: whether the theory was wrong or the questionnaires were.
2.2 What Happened When Somebody Watched Instead of Asking
Dawid Walczak, Radosław Rogoza, and Daniel Jones took an approach that had not been properly tried: they stopped asking people what they are like in general and started asking what they had done that day. They recruited 317 adults in Poland and sent an evening survey to each of them every day for thirty consecutive days, a method psychologists call experience sampling, meaning you catch behaviour close to when it happens rather than relying on somebody’s summary of themselves months later, and each evening’s questions were daily versions of the standard items, so a participant was not reporting whether they are a manipulative person but whether, today, they had done a manipulative thing.
The result is the most useful number in this article. Assessed in the usual way, as traits, the two constructs overlapped by more than seventy per cent, a finding that has driven a decade of argument. Assessed as daily states, across thirty days of actual reported conduct, the overlap fell to roughly sixteen per cent (Walczak, Rogoza & Jones, 2026).
What separated them was the content of the days. The Machiavellian days were days of restraint, best captured by an item about keeping a low profile to get your way, while the psychopathic days were days of exposure, captured by an item about getting into a dangerous situation. The authors’ reading is that the first involves assessing the environment and holding yourself back, and the second involves not doing that.
2.3 The Order of Events
The same study also produced a second result that I think is the more interesting of the two and has received almost no attention, because it concerns sequence rather than size, and the relationship ran in only one direction. A day when somebody behaved more manipulatively than usual predicted a following day when they behaved more impulsively than usual. The reverse did not hold, so an impulsive day predicted nothing about manipulation the day after.
I would be careful about what that licenses, because a pattern across days in 317 people is not a theory of anybody’s life. What it does suggest, and what fits everything in Chapter 5, is that the restrained version is the more stable state and the exposed version is what happens when the restraint has been running for a while and something gives, and that is a rather different picture from the one in which the calculating person simply calculates indefinitely.
2.4 What This Does Not Show
It does not show that the two traits are unrelated, because sixteen per cent of anything is not zero, and the study found the overlap rather than abolishing it. Nor does it show that the daily measure is the right one and the trait measure the wrong one. Both are self-reports, both are subject to the same problem of people describing themselves generously, and a thirty-day study of adults in one country is not a settled account of a construct, and the honest summary is that two ways of asking the same people about the same thing produce answers that disagree by a factor of four, and that the way which disagrees with the received picture is the one that watched for longer.
Evidence status: evidenced. One study, published in a peer-reviewed journal, had 317 participants across thirty consecutive days, with the trait-level comparison reproducing what two prior syntheses had already found. The one-directional sequence result is a single finding in a single sample and is the part I would hold most loosely.
Chapter 3: The Study You Will Be Quoted At
3.1 Three Traits Predicted It and One Did Not
If you have encountered this subject in the last year, there is a reasonable chance you encountered it through a claim about artificial intelligence and cheating, and the claim is worth setting out carefully because it is half right and the wrong half travels further. Rongjian Sun and four colleagues surveyed 812 students at Taiwanese universities, 362 of them undergraduates and 450 postgraduates, and asked about four traits rather than the usual three, adding everyday sadism to the standard set, and they then measured the extent to which each student reported misusing generative AI on their coursework.
Three of the four traits predicted it. Narcissism, psychopathy and sadism each showed a clear and roughly equal association with generative AI misconduct. Machiavellianism showed none (Sun, Tang, Zhou, Loan & Wang, 2025).
That is a striking pattern, and the interpretation that has attached itself to it is equally striking and considerably less defensible: that the calculating are simply too careful to risk it, because they run the odds on being caught and the odds are bad.
3.2 What the Authors Actually Said About the One That Did Not
The authors do not say that, and I think anybody quoting the study should read what they do say, because it is considerably more interesting. They offer two explanations for the empty finding and neither is a claim about restraint. The first is that self-reported unethical conduct is prone to under-reporting, and prone to it most among exactly the people best equipped to conceal what they have done, which makes an honest measurement of this particular group harder than of any other group anybody might want to study. The second is a straightforward admission about their own instrument: in the Traditional Chinese version they used, five items of the Machiavellianism scale did not hang together properly with the rest of it.
Read those two together and the finding changes character entirely. A finding of nothing, produced by a measure that partly failed, taken from people asked to describe their own concealed behaviour, among a group whose defining characteristic is concealment, is not evidence that the behaviour did not happen, and it is a measurement telling you it could not see.
I want to be fair to the study, which is transparent about all of this and which nobody should read as sloppy. The three positive results stand on their own, and the sadism result in particular is worth having, because everyday sadism is measured in a fraction of the work in this area and this is one of the very few places anybody has included it. It is the fourth result, the empty one, that cannot carry the weight put on it.
3.3 A Different Study Found the Opposite
The most economical way to see that is to look at a study that asks a close variant of the same question and reaches the opposite conclusion. Tobias Greitemeyer and Andreas Kastenmüller surveyed 283 people about their willingness to pass off chatbot-generated text as their own work and found six personality predictors. Lower honesty-humility, lower conscientiousness and lower openness on one side, and higher Machiavellianism, narcissism and psychopathy on the other (Greitemeyer & Kastenmüller, 2023). The trait that dropped out of the Taiwanese study did not drop out of this one.
The paper also makes a second finding that I would rather lead with than the first, because it is more useful and gets quoted far less. The strongest predictor in the whole set was none of the dark traits. It was honesty-humility, specifically its fairness component: the extent to which somebody is committed to treating other people equitably whether or not anybody is watching them do it. The dark traits told you something; the ordinary decency measure told you more.
3.4 What Twenty-Three Studies Say Together
Ben Jones and Lee Jones reviewed the whole literature on dark traits and academic dishonesty, including only studies that measured at least two of the traits at once, on the reasonable ground that a study measuring one in isolation cannot tell you which of them is doing the work. Twenty-three studies survived that filter (Jones & Jones, 2026).
Their conclusions run in four directions and none of them is the popular story. Psychopathy showed the most consistent pattern across every form of misconduct they examined, which makes it the reliable trait in this territory rather than Machiavellianism, and Machiavellianism itself was associated with plagiarism and with cheating while being not consistently associated with contract cheating, meaning paying somebody else to produce the work.
Narcissism was weaker and depended heavily on context, and everyday sadism has been examined in so few studies that the authors treat their own conclusion about it as tentative. The contract cheating result is the one I would sit with, because it is the closest thing in this literature to a direct test of the restraint hypothesis and it points the other way. Paying somebody else to write your essay is the version of the offence with the most durable evidence trail and the greatest number of people who know, which is precisely the version a careful calculator ought to avoid most strongly, and it is the version the trait fails to predict.
3.5 Why This Is the Most Useful Mess in the Article
I could have written this chapter as a single clean finding and I have deliberately not, because the disagreement is what a reader can actually use. —————————————————————————————————————————————————————————————————————————————————————————————————————— Evidence status: contested. Three peer-reviewed records on closely related questions, two of which disagree about the direction of the association and one of which finds it depends on which form of misconduct you count. The disagreement changes what a reader should do, which under this project’s rules is the test for whether it deserves space, so it gets a chapter rather than a footnote.
Here is what it changes. If you were hoping to identify the calculating colleague by finding the rule they broke, this chapter is the evidence that the approach does not work even under laboratory conditions, with a captive sample, an unambiguous offence and researchers actively looking. Three competent studies pointed three different ways.
And if you were hoping to screen for the trait, the strongest single predictor in the one study that measured it was not a dark trait at all. It was a measure of ordinary fairness, which is available, well validated, far less legally exposed, and pointed at what you actually care about.
PART II. WHY IT RUNS THIS WAY
Chapter 4: The Questionnaire That Cannot Tell Two Things Apart
4.1 A Distinction Without a Difference
Chapter 2 said the field has spent a decade arguing about whether this trait exists separately from psychopathy. This chapter is where that argument came from, and I ask you to read it as a chapter about your organisation rather than about instruments, because the failure it describes is the one your appraisal system makes.
Joshua Miller and four colleagues put the question directly in three studies, comparing what the standard Machiavellianism measures actually predict against what the standard psychopathy measures predict, across personality ratings and a range of psychological outcomes. Their conclusion was blunt, and it is quoted more often than it is absorbed: the two produced nearly identical profiles, so that on the evidence the measures appear to be assessing the same thing, and the Machiavellianism assessments fail to capture the construct their own theory describes (Miller, Hyatt, Maples-Keller, Carter & Lynam, 2017).
The specific failure is worth stating, because it is not a general vagueness. Theory says a Machiavellian person is controlled and forward-looking. The questionnaires measuring Machiavellianism correlate positively with impulsiveness, the opposite of the trait as described, and the signature of the neighbouring construct.
So the instrument that the whole applied literature has run on for thirty years has been quietly measuring the wrong half of the pair.
4.2 The Overlap, and What Happened When Somebody Tried to Remove It
Michael Grosz and four colleagues set out to fix that rather than to complain about it, and their paper is the best evidence I know of that the theory was fine and the questions were not. They begin from the size of the problem. Two separate syntheses of prior research put the average uncorrected overlap between existing Machiavellianism measures and existing psychopathy measures at .58 and .52, figures that, for a pair of scales supposed to measure different things, are close to what you would expect from two versions of the same test.
They then built two short scales of seven items each, keeping only items that behaved as the theory predicted: Machiavellianism items had to be unrelated to antisocial impulsiveness, while psychopathy items had to be related to it. Across three studies and 4,607 participants, the resulting pair correlated with each other at between .23 and .25 (Grosz, Harms, Dufner, Kraft & Wetzel, 2020). That is the whole argument in two numbers: roughly half the overlap that had convinced people the two traits were one thing was an artefact of which questions somebody had happened to ask, and it went away when somebody asked different ones.
4.3 What This Does to Everything You Have Read on the Subject
I will state the consequence plainly, because most treatments of this topic bury it in a methods paragraph, and it deserves better.
Much of the applied research on Machiavellianism at work used instruments Miller and colleagues describe as measuring psychopathy. When a study reports that Machiavellianism predicted some workplace outcome and used a standard measure, it is not possible to say from the study alone whether the finding belongs to the trait it names or to its neighbour. That does not make the findings false. It makes their labels uncertain, which is a different and considerably more awkward problem, because a false finding can be corrected once somebody notices it while a mislabelled one propagates quietly into everything written afterwards.
This is why Chapter 2 matters more than its sample size would suggest. The daily study is one of the few designs that could separate the two without relying on questionnaires that cannot, and it did.
Evidence status: evidenced. Two peer-reviewed programmes, one comparing what the measures predict across three studies and one rebuilding the measures across three more with 4,607 participants, reaching compatible conclusions from opposite directions. What remains genuinely open is how much of the applied literature is affected, which nobody has quantified and which I cannot state.
4.4 Why the Instrument Fails in the Same Direction Your Organisation Does
This observation is what made me want to write this article rather than the one about detection I had planned. Questionnaires cannot detect Machiavellianism because they take a single reading. A person answers thirty questions on a Tuesday and the answers are treated as a description of how they operate over years, and that method can detect a disposition which expresses itself constantly and visibly, which is why it works tolerably well for psychopathy and for the loud half of narcissism, and it cannot detect a disposition whose entire content is the calibrated decision not to act until acting is safe.
Your organisation has exactly the same instrument. A grievance process takes one reading, triggered by one event, and assessed against whether that event on its own crosses a threshold; an exit interview takes one reading; and a performance review takes one reading a year and asks a single manager to summarise a whole person.
Each of those is well designed to catch a discrete act and structurally blind to a pattern of very small acts, none of which crosses any threshold and all of which point in the same direction. I would put it like this, and it is as close to a thesis as this article has. The calculating colleague is not beating your system. A system built for a different problem is processing them correctly.
Evidence status: practitioner judgement. The comparison between the measurement failure and the organisational one is mine. Each half is evidenced separately and the parallel between them is an argument rather than a finding, and nobody has tested it.
Chapter 5: Camouflage
5.1 The Ones Who Get Caught Are the Clumsy Ones
So far, everything has been about why this is hard to see. This chapter is about the one research programme that has managed to see it, and about the deeply awkward thing it found.
Gerhard Blickle and six colleagues ran two studies with a combined 1,438 participants, using both self-reports and ratings from the participants’ actual colleagues, and asked what happens to somebody high in this trait as their tenure lengthens. Their starting idea comes from biology rather than psychology: a trait can be masked by a skill, much like a harmless species can be masked by a resemblance to a dangerous one.
The skill in question is political skill, meaning a well studied cluster of abilities covering reading social situations accurately, influencing people, building networks and appearing sincere while doing all three; it is measured routinely, it is taught, and it is not itself a dark trait. The first study found the masking effect they predicted. When political skill was high, colleagues gave people high ratings on career role performance, whether or not the person was also high in Machiavellianism. When political skill was low and Machiavellianism was high, ratings were lower, especially among people who had been there a long time (Blickle et al., 2020).
Read that carefully, because it inverts the natural assumption. Time does not expose this trait. Time exposes the people who lack the skill to carry it.
5.2 Reputation Is the Route, and It Can Be Closed
Four years later, three of the same research group asked why the applied literature on this trait and job performance had never settled, with some studies finding a cost and others finding none, and they proposed that everybody had been looking for a direct link where there was an indirect one.
Bastian Kückelhaus, James Meurs, and Gerhard Blickle studied 550 employees rated by 1,127 of their coworkers, using a newer measure designed to capture the strategic content the older ones miss. They found a chain, not a link. People high in the trait developed poorer reputations among their colleagues, and the poorer reputation, not the trait itself, produced the lower assessments of their job performance. Political skill broke the chain: those who had it did not develop the reputation and therefore did not take the hit (Kückelhaus, Meurs & Blickle, 2024).
I find this the single most practically useful result in the literature, and it is worth being precise about why. It identifies the only channel through which this trait has been reliably shown to cost anybody anything at work, which is what colleagues come to think, and it tells you that the channel is closeable.
Evidence status: evidenced. Two multi-source programmes from the same research group, 1,438 participants in the first and 550 rated by 1,127 colleagues in the second, using colleague ratings rather than self-reports for the outcome. Both are from one group and one country, which is the limitation I would attach rather than the design, which is unusually good.
5.3 The Damage Happened Anyway
A second study in the Blickle paper stops this chapter from becoming an argument that political skill is simply a good thing. After finding that political skill masked the trait in colleague ratings, the researchers examined whether it also reduced the conduct. It did not. Employees high in Machiavellianism and high in political skill were reported to engage in harmful interpersonal behaviour and extract organisational resources, and they did so early in the job, when their ratings were most favourable.
So the mask did not change the behaviour. It changed only who could see it, and it worked best when the least was known about the person.
That is the mechanism behind all three rooms in Chapter 1. Nobody was fooled by a performance in the sense of being taken in by an act. The colleagues in that study were rating what they could observe, honestly and competently, and what they could observe had been shaped.
5.4 What a Clean Record Is Actually Measuring
Put the last two chapters together and you get a conclusion I would like you to be able to state in one sentence at work, because it is the sentence that changes a conversation. An organisation’s misconduct record reflects detected misconduct, and detection is not distributed randomly across people. It is concentrated among those who act impulsively, act visibly, and lack the social skill to manage the aftermath. On the evidence in this chapter, the trait that most reliably survives an organisational record is the one defined by not doing those things.
So the absence of a file about somebody is genuinely evidence, and it is evidence of a much narrower thing than it is usually taken for. It is good evidence that this person has not done anything crossing a formal threshold in a visible way. It is very weak evidence about what they have done.
I would separate that carefully from the thing it is not, because the distinction is the whole difference between a useful idea and a licence for suspicion, and nothing here says that a person with a clean record is probably guilty of something. Most people with clean records have clean records because they have behaved well, and this argument does not become evidence about anybody in particular, and it is a statement about what your evidence system can and cannot demonstrate, which is a statement about the system.
Chapter 6: They Are Already Managing You
6.1 Eight Thousand Records and One Uncomfortable Pattern
The natural hope, and it is a hope I have heard expressed in a great many boardrooms, is that this sorts itself out over a career, because people notice, and the calculating do not reach the top. Mehrzad Baktash and Uwe Jirjahn tested that on a German household panel study, using 8,587 observations drawn from 4,631 employed people over the age of twenty-five across three survey waves. Machiavellianism was measured with three items about manipulating, deceiving, and flattering others to get one’s way.
Higher scores were associated with a greater probability of holding a management position, and the relationship was monotonic, meaning it did not level off: those at the very top of the distribution were the most likely to be managers, and those at the bottom the least (Baktash & Jirjahn, 2026). The pattern held for men and for women alike, with no difference between them in the strength of the association.
The effect is real and it is not enormous, and I would rather give you the size than let the direction do all the work. About four in every hundred observations in that sample were of somebody holding a management post, and one further point on the five-point Machiavellianism scale went with roughly one and a third more managers per hundred, which is a substantial proportional increase applied to a small base.
6.2 Which Way the Arrow Runs
The interesting question with any finding of this shape is whether the job produced the personality or the personality produced the job, and the authors took some trouble over it. Their reading is that the arrow runs from the trait to the position rather than the other way round, meaning that people higher in the trait are more likely to become managers, rather than that becoming a manager makes people more Machiavellian.
I would hold that a little more loosely than they do, because establishing direction in survey data is difficult even with good tools. But it is the more plausible of the two on general grounds, since the trait is measured as a stable disposition and a management appointment is an event.
6.3 What This Does Not Licence
This is the point in an article of this kind where a reader reasonably starts drawing conclusions about their own manager, and I would ask you not to. A pattern across thousands of people tells you nothing whatever about one person, and the base rate makes that concrete: the overwhelming majority of managers in that sample were not high in the trait, because the overwhelming majority of people are not. The finding shows that selection processes do not filter this out, and that a system many people assume is quietly self-correcting is not correcting in this direction at all.
There is one further result I want to mention and then set aside, and the same two researchers have a more recent paper reporting that the association with reaching top-level management holds for men and not for women, which is an interesting and troubling claim. It exists at the time of writing as a working paper rather than as a peer-reviewed publication, so under this project’s citation rules it does not enter the argument and I record it here only so that you are not surprised to meet it.
6.4 The Damage Is Ordinary and It Is Constant
Nothing in this article should leave you with the impression that this trait is harmless to organisations because it is hard to see. Mirko Duradoni and five colleagues reviewed twenty-one studies on the darker traits and the general category of workplace conduct organisations least want: theft, sabotage, withholding effort, undermining colleagues, and the rest of it. Machiavellianism and psychopathy both showed positive associations with both the organisational and interpersonal forms, with Machiavellianism ranking second among the three traits examined, behind psychopathy alone (Duradoni, Gursesli, Martucci, Gonzalez Ayarza, Colombini & Guazzini, 2025).
That review carries the measurement problem in Chapter 4, since most of its constituent studies used instruments that cannot separate the two traits, and I would read its Machiavellianism column with that firmly in mind. What survives the qualification is the direction, which is consistent, and the aggregate picture, which is of an ordinary and continuous cost rather than a dramatic and occasional one.
Which is the last piece of the shape. The harm here is not an incident. It is a rate.
PART III. BEFORE YOU USE ANY OF THIS
Chapter 7: The Five People You Will Mislabel
Everything so far describes one pattern. This chapter is about the five people who resemble it and are not it, and I would say the risk of getting this wrong is higher here than anywhere else in the series, for the reason Chapter 1 gave: with this trait you will never have the behavioural evidence that would correct you.
If reading this article makes you more willing to apply the word to somebody, it has failed, and I would rather say that at the top of the chapter than at the bottom of it.
7.1 The Politically Skilled One Who Is Not Working You
This is the important one, and I have put it first because Chapter 5 has just spent two thousand words explaining that political skill hides the trait, which makes it very easy to slide into treating the skill as the evidence. It is not, and the studies make that explicit. Political skill is a set of abilities: reading a situation accurately, influencing people, building useful relationships and coming across as sincere, and it is measured separately from Machiavellianism in every one of those studies precisely because the two are different things, and the whole design of the masking research depends on people having one without the other.
The colleague who is unusually good at reading a room, who knows who to talk to before a meeting, and who can carry a difficult proposal without a fight is showing you a skill that most organisations undervalue and many people wish they had. What Chapter 5 says is that if somebody also has the trait, this skill will conceal it. It does not say, and cannot say, that the skill indicates the trait.
The practical test I would offer is about who benefits. Political skill deployed to get a decision made, a project unstuck or a quiet person heard looks identical, moment to moment, to political skill deployed to move a cost onto somebody else. Over a year, the beneficiaries differ, and the pattern of who ends up carrying things is the only thing separating them.
7.2 The Impulsive One Everybody Calls Calculating
Chapter 4 is the reason for this one, and it applies to observers as much as to instruments. When somebody does something clearly self-serving and clearly harmful, the word that arrives is usually calculating, because harm looks intentional and intention looks planned. Everything in this article argues the opposite, and the visible, traceable, discoverable act is the signature of the impulsive trait rather than the strategic one, and the thirty-day study found exactly that separation: the dangerous situations belonged to the psychopathy days, not the Machiavellianism ones.
So the colleague who blew up a meeting, sent the email everybody has now seen, or took credit for something in front of the person who did it is showing you something, and on this evidence what they are showing you is poor impulse control rather than a long game. Calling that Machiavellian is not merely imprecise. It reverses the diagnosis.
7.3 The One Who Wants to Be Admired, Not to Win
Grandiose narcissism and Machiavellianism both produce self-serving behaviour and are routinely used as synonyms in ordinary speech, and the goal underneath them is not the same one. Narcissistic behaviour is organised around standing: being seen as exceptional, admired, deferred to. It is therefore performed in front of people, because an audience is the point, and it collapses in private. Machiavellian behaviour is organised around outcome, and an audience is a risk rather than a reward, which is why the daily item that captures it is about keeping a low profile.
The practical difference is that one will tell you what they did, and the other will make sure the question doesn't arise. If your colleague’s difficult behaviour is in front of an audience, you are almost certainly reading the wrong article in this series, and the one on grandiose narcissism is the right one.
7.4 The Private One
A fourth kind of colleague fits the shape of Chapter 1 exactly and does nothing at all, and I mention them because the pattern this article describes is defined partly by absence, and absence has many causes. Somebody who does not share their reasoning, does not socialise with the team, keeps their own counsel about decisions and cannot easily be read is producing the same evidential vacuum as the person in Chapter 1. They may be reserved, may have been burned somewhere previously, may come from a working culture where transparency about one’s own thinking is not the norm, or may simply be uninterested in the currency of workplace disclosure.
The vacuum is the same, and what differs is whether anything is moving inside it. The honest answer is that from the outside, in the short term, you cannot tell, which is not a comfortable thing to write in an article about detection and is the accurate thing.
7.5 The One Who Needs a Clinician
This boundary is categorical rather than a matter of degree and it works the same way here as in every other article in this series. Machiavellianism as studied here is a continuous trait on which every person sits somewhere, most of us near the middle, and it is not a clinical category at all. There is no diagnosis called Machiavellianism, which distinguishes it from narcissism and from psychopathy, both of which have clinical relatives frequently confused with the everyday trait.
A score is not a diagnosis, an observation is not a diagnosis, and nothing here can be turned into one about anybody. Where somebody’s conduct at work suggests they are unwell rather than difficult, that is a matter for occupational health and a qualified clinician with direct access to the person, and not a matter for a colleague with a reading list.
7.6 And the Numbers Do Not Describe a Person Anyway
The general caution applies here with more force than usual, and I would rather state it than let the previous six chapters imply otherwise. Every figure in this article states how measurements relate to one another across large groups of people. None of them is a statement about an individual, and the step from one to the other is not small or merely technical. It is the difference between knowing that taller people are on average heavier and knowing anything at all about the weight of the next person you meet.
Three shorter cautions belong beside it. Nobody knows the base rate of this trait in any workplace population, so treat any claim about the percentage of executives who are Machiavellian as unsupported until you have seen where it came from. Almost all of the evidence here comes from Europe, Taiwan and North America, and the samples in the two most load-bearing chapters are Polish and German. And almost none of it establishes cause, with the panel study in Chapter 6 making the only serious attempt at it in this whole article.
PART IV. WHAT TO DO
Chapter 8: Four Moves and Three Refusals
Evidence status: mixed and stated per item. This is the weakest section of the article and I would rather say so at the top than let the confident tone of a numbered list carry an authority the evidence does not. Nobody has run a trial of any of this. Each item below carries its own status, and two of them rest on nothing more than a finding pointed in a useful direction.
8.1 Count Episodes, Not Incidents
This follows directly from Chapter 4, and it is the only move here that I think genuinely changes outcomes. Your organisation asks, of each event, whether that event on its own is serious enough to act on. That question has a correct answer and it is almost always no, which is why the file stays empty. The question worth asking instead is how many times something small has happened, over what period, and whether the small things point the same way.
In practice, that means keeping a dated note of things that are individually not worth raising: the commitment made in a meeting but not in writing, the decision that arrived already taken, the information that reached you late and reached somebody else on time. Not as a dossier about a person, which is both unpleasant and legally unwise, but as a record of events with dates attached, which is what turns an impression into something a third party can assess.
Iwould add two honest warnings to that. It is slow, because a rate needs months before it is a rate. And it is the kind of activity that looks bad if it is discovered and cannot be explained, so the note should contain only things you would be content to read aloud. Practitioner judgement, reasoned from the measurement argument at 4.4.
8.2 Ask the Colleagues, Not the Manager
This is the most directly evidenced item on the list and it is aimed at organisations rather than at individuals. Chapter 5 identified the single channel through which this trait has been shown to cost anything: what colleagues think. Both of the multi-source studies found it in coworker ratings, and neither found it in anything a manager could see on their own, which is unsurprising once you notice that a manager is the audience the behaviour is organised around.
So any process that collects a view from several colleagues, independently, and over time, is looking in the one place the evidence says something is visible. Multi-rater feedback is the obvious instrument, and I would use it with two adjustments: ask about specific behaviour rather than qualities, and compare across cycles rather than reading each round on its own, because a rating is what you are trying to see.
This is worth doing, even though it is expensive, because it is the only recommendation here with an actual finding behind it. Evidenced as to the channel; practitioner judgement as to the instrument.
8.3 Put the Cost Where the Decision Is Made
Chapter 5 also tells you why the ordinary levers fail. The behaviour it describes is not rule-breaking, so a rule is the wrong tool.
The pattern comes from a structure in which the benefit of a decision lands on the person taking it and the cost lands somewhere else, later, on somebody who was not in the room. Any change that reconnects those two things reduces the return on the behaviour without anybody having to be accused of anything: decisions recorded with the name of the person who took them, commitments carried forward into the following meeting rather than being allowed to expire, and the person who will bear the consequence present when the decision is made.
It is the dullest recommendation in this article and it is the only one an organisation can act on unilaterally, without evidence about any individual, and without a confrontation. Practitioner judgement. The underlying mechanism is evidenced; this design response is not tested.
8.4 Write Down What You Agreed, on the Day
This is the small personal version of 8.3, and I include it because it is the thing I have seen work most often in practice. A short note sent afterwards, saying what was agreed and by whom, costs a couple of minutes and converts an oral understanding into a dated record. It is not aimed at catching anybody and it should not read as though it is. Its value is that it removes the ambiguity the pattern depends on, and it does so without requiring you to have a theory about the other person or to say anything about them at all.
The reason it works, on the argument in this article, is that this is the one trait whose operation requires the absence of a record, and a record is cheap. Practitioner judgement, and the most-used item on this list in my own advisory practice.
8.5 Do Not Test Candidates For It
Instruments exist; some are sold as integrity or derailer assessments rather than under any dark trait name, and I would not use one for this purpose. Four reasons, briefly. Chapter 4 establishes that the standard measures do not reliably measure the thing they name, so a score is ambiguous before anybody has answered a single question, and the measures are in any case self-reports of concealment, which is the design problem Chapter 3 watched defeat a competent research team with funding and consent behind it.
Screening candidates on a quality this loaded also carries real legal exposure and no demonstrated benefit. And on the one occasion in this article where somebody measured an alternative, the alternative won: honesty-humility, and specifically its fairness component, out-predicted every dark trait in the study that included it.
If you want to assess something in this territory, assess that instead. It is a mainstream personality dimension; it is not a diagnosis of anybody, and it is pointed at what you actually care about. Evidence as to the measurement failure; practitioner judgement as to the recommendation.
8.6 Do Not Treat a Clean Record as Evidence
I mean this in both directions, and the second direction is the one people forget. A clean record is weak evidence that somebody has behaved well, for the reason set out at 5.4. It is also not evidence that they have behaved badly, and an argument of the form there is nothing on file, which proves how careful they are is unfalsifiable and should be treated as such wherever you meet it, including when you are the one making it.
The correct handling of an empty file is to say that it is empty and that this settles less than people think. Practitioner judgement, reasoned from Chapter 5.
8.7 Do Not Set Out to Catch Them
This is the refusal I would most like to be taken seriously, and it is aimed at the reader who has recognised somebody in Chapter 1 and is now feeling energetic. Chapter 3 watched three research teams, with captive samples, unambiguous offences, ethical clearance and funding, fail to establish something as basic as whether this trait predicts cheating on an essay. You are no better positioned than they were, and the attempt has a specific failure mode: you become the person with the theory about a colleague, and the trait most reliably associated with the behaviour you describe is not one anybody will measure.
Everything useful in this chapter is a change to a system or a habit, and none of it requires a conclusion about a person. That is not squeamishness. It is where the evidence happens to be. Practitioner judgement.
Chapter 9: What This Changes
9.1 The Short Version
If you read nothing else here, read this. The trait is restraint, and restraint leaves no evidence. Machiavellianism as its own theory describes it is strategic self-interest exercised with control over time, which means the behaviour it produces is calibrated to stay below whatever threshold would trigger a response. An organisation’s record reflects things above that threshold.
The instruments share the same blind spot as your organisation. Standard measures of the trait correlate with impulsiveness, the opposite of the theory, and two syntheses put their overlap with psychopathy measures at .58 and .52. Rebuilt properly, the overlap falls to about a quarter of what it was, which suggests the theory was sound and the questions were not.
Watching beats asking, and it is not close. Across 317 people over thirty consecutive evenings, the overlap between the two traits fell from over seventy per cent at the trait level to about sixteen at the level of what people actually did. The Machiavellian days were the ones spent keeping a low profile.
People who get caught are those without the social skill to avoid it. Across 1,438 people, high political skill produced high colleague ratings whether or not the person was also high in this trait, and the trait cost people only when the skill was absent. In the second study of that same programme, the skilled ones were doing the harm anyway.
Which means the empty file tells you about your system rather than the person. It is good evidence that nothing crossed a formal threshold visibly, and weak evidence about anything else, and the two are constantly confused.
9.2 So What Do You Do
Three things, and not one of them is a conversation about somebody’s character. Count over months, not events. An event will always be too small to act on, because being too small to act on is the design. A dated note of small things, kept honestly and containing only what you would read aloud, converts an impression into something assessable.
Ask the people alongside them, repeatedly. Colleague ratings are the only place this has been shown to be visible, and they show change over cycles rather than as a finding in any one of them. Reconnect the decision to its consequence. When the person making the decision is also the person carrying it out, most of the return on this behaviour disappears, and no accusation is required to bring that about.
9.3 Why Any of This Matters
Two reasons, and the second is the one I would leave you with. The first is that the usual advice on this subject is to become a better observer, and it is close to useless, because the problem is not the quality of your observation. Three of the best research groups working on this could not establish, with instruments and funding and consent, what you are being told to determine by paying closer attention in meetings. You are not failing at something achievable.
The second is more uncomfortable and it is about organisations rather than about people. Every system described in this article behaves correctly: the grievance process is right to require an event, and the appraisal is right to ask a manager, and the complaint threshold exists to protect people from exactly the kind of accumulated impression this article has spent a chapter explaining how to build.
That protection is worth having and I would not remove it.
Which leaves a genuinely awkward conclusion rather than a satisfying one. The blind spot is not a fault in the system. It is the cost of the system’s virtues, and the only honest responses to it are the slow, structural, unexciting ones in Chapter 8. Anybody selling you a faster answer is selling you a way of being confident about a person, which is the one thing none of this evidence supports.
9.4 What I Still Do Not Know
Four things remain genuinely open to me and I would rather state them as questions than resolve them by leaving them out. How much of the applied literature on Machiavellianism at work is actually about psychopathy? Chapter 4 establishes that the question is real and nobody has quantified the answer, which means every figure in Chapter 6 carries an uncertainty I cannot size.
Whether the thirty-day separation reproduces. It is one study in one country; it is the load-bearing finding in this article, and I wrote the article around it because it is the best design anybody has applied to the question, not because a single study should carry that weight.
Whether the AI misconduct picture resolves in either direction, or whether self-report is simply the wrong method for this particular question and the disagreement in Chapter 3 is permanent.
And whether any of Chapter 8 works, since none of it has been tested against this problem and two items rest on my own practice rather than on anything published.
A closing note on how I have handled the material. This is the dark trait with the widest gap between what is confidently asserted about it and what has actually been shown, and the gap is unusual in that it runs through measurement rather than interpretation. I have tried to keep the strong findings and the weak ones visibly apart, to mark my inferences as mine, and to report the results that spoil the argument alongside the ones that make it, of which Chapter 3 is entirely composed. What I would most like to have made harder is a confident verdict about the colleague you had in mind while you were reading, and I know an article about invisible behaviour is unusually easy to read as confirmation.
