PubMed, PsycINFO, Embase, Scopus, and Web of Science were searched from inception to 2 September 2025. We combined controlled vocabulary and free-text terms related to Dark Triad traits and workplace bullying, including synonyms for each trait and multiple operationalizations of bullying (e.g., abusive supervision, harassment, workplace aggression). Reference lists of eligible studies were screened manually, and citation chaining was used to capture additional records. Expert consultation was undertaken prior to finalizing the search strategy. This involved consultation with a health sciences academic librarian and two researchers with expertise in occupational psychology and workplace bullying, who reviewed the draft search strings, controlled vocabulary terms, and a set of sentinel papers to assess search sensitivity and coverage. Feedback from this process informed refinements to trait-related and outcome-related keywords, including alternative operationalizations of workplace bullying (e.g., abusive supervision and interpersonal counterproductive work behavior).

Studies were included if they: (1) reported quantitative associations between one or more Dark Triad traits (Machiavellianism, narcissism, psychopathy, or composite indices) and workplace bullying perpetration; (2) involved working-age participants in occupational settings; and (3) provided sufficient data to calculate or extract effect sizes. We excluded reviews, non-English papers, and studies focusing solely on victimization. To ensure our analysis focused on direct, predictive associations, studies that examined Dark Triad traits only as mediators or moderators were excluded from the meta-analysis but included in the narrative analysis.

Records were screened using Rayyan, a web-based platform designed to support systematic review screening, by two reviewers (SS and LY), who independently assessed titles, abstracts, and full texts for eligibility. Titles and abstracts were reviewed first, followed by full texts of potentially eligible articles. Disagreements were resolved through discussion or third-party arbitration (SM). Reasons for exclusion at the full-text stage were documented systematically (Appendix Table S1). Extracted data (Appendix Tables S2a,b) included study characteristics (country, year, sample size, occupational sector), trait measures [e.g., Short Dark Triad (SD3), Dirty Dozen], bullying outcomes (e.g., Negative Acts Questionnaire, Counterproductive Work Behavior) – Individual (CWB-I) indices, effect estimates, funding, and conflicts of interest. Where data were missing or unclear, authors were contacted for clarification. The extraction process was piloted to ensure consistency across reviewers.

Risk of bias was assessed using an adapted Joanna Briggs Institute (JBI) checklist (Joanna Briggs, 2017; Moola et al., 2020) for analytical cross-sectional studies, modified for occupational-personality research to cover: sampling adequacy; validity and reliability of trait and bullying-perpetration measures (including alignment with construct definitions); handling of confounders; appropriateness of analytic methods; treatment of missing data; mitigation of common-method bias (e.g., multi-source or time-lagged designs); and transparency of reporting/ethics. The checklist evaluates key domains including sampling adequacy, validity and reliability of exposure and outcome measures, identification and control of confounding variables, appropriateness of statistical analyses, and transparency of reporting. Given the predominance of occupational and personality-based survey research, additional attention was paid to common-method bias, reliance on self-report measures, and temporal separation of variables where applicable. Overall risk-of-bias judgements were derived using prespecified criteria across domains. Two reviewers assessed each domain independently and resolved disagreements by consensus after a calibration exercise. Overall ratings followed a prespecified rule: “Low risk” only if all critical domains were adequate; “High” if ≥3 critical domains were inadequate or any fatal flaw was present; otherwise “Some concerns.” All included studies were rated “Some concerns.” Full domain-level judgements are provided in Appendix Table S3.

All statistical analyses were performed in R (Version 4.4.1; R Core Team, 2024) using the ‘metafor’ package. We provide the meta-analytic dataset Appendix Table S4 to enable independent re-analysis; code is available on request. The primary outcome was the correlation between Dark Triad traits and workplace bullying perpetration. When multiple measures were reported, we prioritized validated scales of perpetration. Effect sizes were standardized to correlation coefficients using Fisher’s z transformation. Random-effects models were applied using restricted maximum likelihood in R. Heterogeneity was assessed with Q, I², and τ² statistics, which tests whether observed variability exceeds that expected by chance; the I² statistic, which estimates the proportion of total variance attributable to between-study heterogeneity; and τ², which represents the estimated between-study variance in random-effects models. Subgroup analyses examined outcome subtypes (bullying perpetration, abusive supervision, interpersonal deviance), trait facets, and measurement instruments. Where data permitted, exploratory subgroup and sensitivity analyses examined whether pooled associations differed according to the Dark Triad measurement instrument used (e.g., Short Dark Triad, Dirty Dozen). Sensitivity analyses also excluded high-risk studies or outliers to test robustness. Funnel plots and Egger’s tests assessed small-study effects, with recognition of limited power when k < 10. Certainty of evidence was appraised using the Grading of Recommendations, Assessment, Development and Evaluations approach (Guyatt et al., 2008), considering study quality, consistency, directness, precision, and publication bias.

The database search identified 89 records, with 16 meeting the inclusion criteria (Figure 1). Nine contributed quantitative data to the meta-analysis; seven were excluded due to insufficient data, a non-quantitative design, or a focus on mediators or moderators (Figure 1). The included studies (Table 1) were drawn from diverse regions, including Europe, North America, and Asia, and were conducted in various settings such as healthcare, corporate, and education. Sample sizes ranged from under 100 to over 1,800 participants. Eight studies were cross-sectional surveys, although the rest used longitudinal or experimental designs. Dark Triad traits were assessed using validated instruments like the SD3 and Dirty Dozen, while bullying outcomes were measured with validated indices or adapted questionnaires.

Across included studies, there was variability in how dark personality traits were operationalized. Some studies assessed all three Dark Triad traits concurrently, whereas others examined only one or two traits in isolation. A small number of studies additionally incorporated broader dark personality frameworks, such as the Dark Tetrad, typically including everyday sadism. However, such studies were limited in number and varied substantially in measurement and outcome definitions, precluding quantitative synthesis beyond the Dark Triad traits.

The included studies also differed in methodological design and cultural context. Most employed cross-sectional survey designs, whereas a smaller number used time-lagged, longitudinal, quasi-experimental, or multi-source approaches. Studies were conducted across diverse cultural settings, including North America, East Asia, and Europe, with the United States and Korea most frequently represented. These differences in design and context likely contributed to the substantial heterogeneity observed across pooled analyses.

In Table 1, 13 studies were judged to have “Moderate risk” of bias, largely due to cross-sectional design, potential selection bias, and limited adjustment for confounding. Three were rated as lower risk, particularly those with multi-wave or multi-source designs. None were free of bias in all domains. Detailed assessments are provided in Appendix Table S3.

Pooled analyses indicated that all Dark Triad personality traits were positively associated with bullying perpetration (Table 2 and Figure 2). The composite Dark Triad score showed a large, pooled correlation (r = 0.63 [95% CI 0.32–0.82]; k = 4). Psychopathy yielded the largest individual association (r = 0.53 [95% CI 0.28–0.71]; k = 5). Narcissism and Machiavellianism demonstrated moderate effects (r = 0.40 and r = 0.35 respectively). Substantial heterogeneity was observed across syntheses (I² > 90%). Subgroup analyses showed consistently positive associations across outcome subtypes (bullying perpetration, abusive supervision, interpersonal deviance), though differences between subtypes were not statistically significant (Table 3). Prediction intervals were wide, indicating considerable between-study variation. Studies varied in the instruments used to assess Dark Triad traits, most commonly the SD3 and the Dirty Dozen. However, the number of studies per trait and per instrument was limited, restricting the feasibility and interpretability of formal moderator analyses by measurement scale. Sensitivity checks indicated that the overall direction and relative magnitude of associations were consistent across instruments, although precision was reduced in subsets defined by specific measures.

Beyond pooled estimates, clear patterns emerged across the 16 studies. Machiavellianism was a reliable predictor of bullying, particularly in competitive, market-oriented organizational cultures, while its association with perpetration weakened in more collaborative environments, suggesting strong context dependence. Supervisors scoring higher on psychopathy, as measured using the SD3 in several studies, appeared less constrained by reputational costs, allowing abusive supervision to persist. Narcissism showed more subtle effects. Vulnerable narcissism reliably predicted abusive supervision through maladaptive attributions and shame, whereas grandiose narcissism was less strongly implicated. Contextual moderators further shaped outcomes: for example, abusive supervision provoked retaliatory aggression through injustice perceptions, particularly among employees high in narcissism. Supervisors were more likely to engage in abuse when experiencing identity threat and low moral identity, and situational stressors such as job insecurity combined with leader narcissism to heighten aggression intentions. In all, these findings highlight that dark traits exert their influence within a wider matrix of organizational climate, justice perceptions, and leader-follower dynamics.

Funnel plots (Figure 3) suggested no systematic small-study bias, although statistical tests were underpowered due to small k. Leave-one-out analyses showed stable pooled estimates, and alternative random-effects estimators produced results consistent with DerSimonian and Laird models (DerSimonian and Laird, 1986), confirming robustness (Table 4).

We observed heterogeneity across traits and outcome subtypes (bullying perpetration, abusive supervision, CWB-I). This likely reflects measurement differences, sampling frames (occupational sectors and countries), and model adjustments. Small-study bias diagnostics (Egger tests) were under-powered and therefore interpreted cautiously. Leave-one-out analyses indicated influence for a small number of effects in the psychopathy and Machiavellianism sets, while narcissism appeared stable under case deletion. Fixed-versus random-effects comparisons suggested that heterogeneity matters, and model sensitivity using Paule-Mandel with Knapp-Hartung adjustments produced conclusions consistent with the primary models. Therefore, these checks increase confidence that the direction and relative magnitudes of the pooled associations are not artefacts of a single estimator or idiosyncratic study.

Differences in study design and cultural context provide important insight into the observed heterogeneity. Cross-sectional designs relying on single-source self-report data may inflate associations through shared-method variance, whereas longitudinal, experimental, or multi-source designs offer stronger temporal and causal inference and may yield more conservative estimates. Cultural context may further shape how dark personality traits are expressed and perceived in workplace interactions, particularly in relation to power distance, authority norms, and reporting practices. Consequently, effect sizes should be interpreted as average tendencies across heterogeneous contexts rather than as uniform estimates applicable to all organizational or cultural settings.

The translation of these findings is organizational, not clinical. Although personality traits should not be used as exclusionary screening tools, awareness of dark personality tendencies may inform leadership development, governance, and prevention strategies. Organizational safeguards, such as clear behavioral standards, robust reporting mechanisms, procedural justice, and a strong psychosocial safety climate, may limit opportunities for dispositional risk to translate into sustained bullying. Interventions that focus on accountability, ethical leadership, and work design may therefore be more effective than approaches that target individual traits in isolation.

Personality assessments should not be used as blunt screening or exclusion tools in recruitment or promotion, but rather as one component of a broader evaluation strategy. Importantly, the persistence of psychopathic leadership behaviors may reflect structural, rather than individual, determinants. Positions that reward dominance and power signaling inherently attract individuals high in callous-manipulative traits. By contrast, rotational or collective leadership models, such as those adopted in some UK universities, minimize such selection pressures by decoupling power from positional permanence. In this view, psychopathy is selected, not created, by systems that conflate authority with personal dominance. Reforms that flatten hierarchies or distribute decision-making power may therefore constitute one of the most effective safeguards against toxic supervision. We recommend integrating these measures with contextual judgement, behavioral feedback, and leadership-development frameworks to mitigate misuse while harnessing their diagnostic value. In addition to, a risk-management approach consistent with occupational health and safety principles is warranted: build personality awareness and behavioral accountability into supervision training, using 360-degree feedback and mentoring to surface blind spots, especially in leaders with grandiose or callous tendencies. Establish system safeguards through zero-tolerance policies, confidential and credible reporting channels, timely investigation protocols, and just-culture responses that separate accountability from blame. Improve work design to reduce opportunities for bullying by clarifying roles, allocating workloads fairly, ensuring transparent decision-making, and embedding procedural justice in performance management. Strengthen the PSC via visible senior-management commitment to psychological health, dedicated resources for prevention, and routine monitoring of interpersonal climate. Additionally, provide support and remediation by offering coaching for perpetrators who can change and ensuring access to evidence-based assistance for targets and teams (e.g., restorative practices, conflict management).

Strengths include prospective preregistration, a comprehensive multi-database search, duplicate screening and extraction, and transparent risk-of-bias procedures adapted from the JBI checklist. We synthesized trait-specific and DT-total effects and examined outcome subtypes to avoid conflating interpersonal aggression with broader counterproductive behaviors.

Limitations warrant restraint in interpretation. The included studies sampled participants from diverse occupational contexts, including healthcare, corporate, education, and mixed-sector workforces. While this diversity reflects the widespread nature of workplace bullying, it limits the generalizability of pooled estimates to specific occupational settings and likely contributes to between-study variability.

Methodological heterogeneity also arose from differences in the operationalization of dark personality traits. Some studies assessed individual Dark Triad traits in isolation, whereas others examined partial combinations or the full triad, which may influence observed effect sizes and comparability across studies. Although broader frameworks such as the Dark Tetrad are theoretically relevant, the limited number and heterogeneity of such studies currently restrict the feasibility of quantitative synthesis or moderator analyses beyond the Dark Triad.

In addition, most included studies were cross-sectional and relied on single-source self-report data, which may inflate associations through shared-method variance and preclude causal inference. The modest number of meta-eligible effects per trait further constrained precision and limited the power of small-study bias diagnostics. Finally, although our focus on the Dark Triad ensured theoretical coherence and methodological comparability, related constructs such as everyday sadism and protective traits (e.g., empathy or honesty–humility) were outside the scope of this review and warrant future synthesis.

Measurement-related limitations should also be noted. Brief instruments such as the Dirty Dozen are widely used in organizational research due to their efficiency, yet their psychometric properties and construct coverage have been questioned. Variation in scale length and dimensional coverage may have contributed to between-study variability in effect sizes. Accordingly, pooled estimates should be interpreted as reflecting broad trait tendencies rather than fine-grained facet-level associations. Future research would benefit from greater consistency in measurement and from using instruments with stronger construct validity when examining dark personality traits in workplace contexts.

To refine estimates and inform prevention, future research should prioritize longitudinal, multi-wave, and multi-source designs to establish temporal order and reduce shared-method bias; report meta-analytic inputs transparently (e.g., r, n, and confidence intervals) and make materials openly available. Trait delineation should also improve, distinguishing fine-grained facets such as grandiose versus vulnerable narcissism; psychopathy’s triarchic dimensions of boldness, meanness, and disinhibition, which map onto primary (bold–mean) and secondary (disinhibited) psychopathy profiles (Patrick et al., 2009; Sellbom and Drislane, 2021); and tactical versus cynical forms of Machiavellianism. The scope should further expand to the Dark Tetrad and to protective dispositions such as empathy, honesty-humility, mentalization, and resilience. Future studies should pre-register and test contextual moderators that determine when traits translate into harm (e.g., psychosocial safety climate, organizational justice, leadership style, power asymmetries, job insecurity, and role ambiguity). Finally, organizational interventions should be evaluated using field or quasi-experimental designs, such as staggered PSC initiatives, leadership feedback and coaching, or changes to accountability and reporting systems, with follow-ups tracking downstream health and productivity outcomes. Importantly, this review does not imply that dark personality traits inevitably lead to bullying, but rather that they represent one component within a broader multilevel system of risk.