# What Drives Evil Acts?

- Link: https://thedarksideofhumanity.com/what-drives-evil-acts/
- Published: 2026-06-14T20:25:19-04:00
- Author: Mandi Martyr

Glossary Case TermsX Moral disengagementCognitive and social processes that suspend self-condemnation, allowing a person to participate in harm while preserving a favourable self-image. DehumanizationTreating people as less than fully human, or as objects, contaminants, numbers, enemies or obstacles. Moral exclusionPlacing a person or group outside the boundary in which fairness, empathy and ordinary obligations are believed to apply. Displacement of responsibilityAttributing one’s own conduct to an authority, rule, chain of command or institutional demand. Diffusion of responsibilityWeakening personal accountability by spreading action or decision-making across a group. Demand characteristicsCues that tell research participants what behaviour investigators expect, potentially distorting the result. Ecological modelA public-health framework that examines interacting individual, relationship, community and societal conditions. Risk factorA characteristic associated with increased probability. It is not a diagnosis, destiny or proof that a specific person will act violently. Criminal Psychology // Human Malevolence What Drives Evil Acts? Cruelty is not a single trait. It is a process of agency, identity, authority, opportunity and permission. The word evil can express moral horror, but it is a poor scientific diagnosis. Severe harm is better examined as conduct produced through interacting levels: an individual makes choices; relationships shape access and loyalty; groups define norms and enemies; institutions allocate power and impunity; and wider systems determine whose suffering is visible, tolerated or rewarded. None of those levels erases responsibility. Together, they explain why cruelty can be planned by committed offenders, normalized inside organizations, carried out by compliant participants, or ignored by bystanders who would reject the same act in another setting. This dossier does not label mental illness, trauma, poverty or biology as “evil.” Most people exposed to those conditions do not become violent. Explanation must never become stigma, prediction must never be confused with certainty, and victims must never become props in a theory about offenders. Institutional space can distance decision-makers from consequences. Photo: Declan Sun / UnsplashCrowds do not erase agency; they alter information, identity and perceived permission. Photo: Erica Wheadon / UnsplashOversight can protect, intimidate or merely document harm after the fact. Photo: Dong Xie / Unsplash 01Research SnapshotThe strongest conclusions and the limits attached to them 4Ecological levelsIndividual, relationship, community and societal conditions interact. The model avoids reducing violence to one defective personality or one bad environment. 8Disengagement mechanismsBandura’s framework identifies eight recurring ways people neutralize self-sanction before, during or after harmful conduct. 0Deterministic markersNo brain scan, gene, diagnosis or childhood event can identify an inevitable violent offender. Risk factors are probabilistic and context-dependent. 65%Classic Milgram conditionIn the best-known baseline condition, 26 of 40 participants continued to the final switch. The result demonstrates pressure, not mindless universal obedience. 6Days at StanfordThe prison simulation ended early, but its methods, coaching, sampling and interpretation prevent it from functioning as clean proof that roles automatically create cruelty. 1Moral agentSystems influence action, but institutions do not act without people. Responsibility may be distributed; it does not disappear. 02Research UpdateWhat changed in the modern reading of classic “evil” experimentsThe contemporary literature is less interested in a theatrical claim that “anyone becomes a monster” and more interested in identification, leadership, legitimacy, gradual commitment and institutional permission. This does not make situations unimportant. It makes the mechanism more precise. What changedRole absorption is no longer enoughReanalysis of the Stanford Prison Experiment documented demand characteristics, researcher involvement and direct encouragement. The dramatic outcome cannot be treated as an untouched emergence of guard behaviour. What it establishedPeople follow identities and leadersLater work emphasizes that participants often comply when they identify with a cause, authority or group project, not simply because they stop thinking. Commitment can produce both abuse and resistance. What remains openPrediction is still weakResearch can identify conditions associated with harm, but individualized forecasts remain limited, especially for rare severe violence. False confidence can produce stigma, surveillance and unjust intervention. Systems often become visible through their corridors, routines and controlled access. Photo: Nguyễn Hiệp / Unsplash 03Interactive Forces MapFive levels that can increase or restrain harmful conduct Level 01Individual Level 02Relationship Level 03Community Level 04Society Cross-cuttingIdentity Individual levelAgency, learning and self-regulationPersonal history, temperament, beliefs, substance use, prior violence, perceived grievance, empathy and self-control can alter risk. None is sufficient alone. The decisive question is how a person interprets a situation and what conduct they choose when restraint, opportunity and social permission collide.ChoiceLearningSelf-control 04Clickable Research ChronologyHow the dominant explanation moved from obedience to identity and systems 1961–63Milgram’s obedience studiesParticipants were pressured to continue a procedure they believed harmed another person.Established: authoritative settings, escalating commitments and scripted prompts can produce high compliance. Limit: participants varied, many resisted, and later scholarship disputes whether obedience was “blind” or based on identification with science. 1963Arendt and the “banality” debateAdministrative normality became a way to discuss extraordinary state crime.Interpretive contribution: atrocity can be organized through routine offices, rules and professional language. Limit: the phrase is often misused to imply perpetrators lack ideology, hostility or ambition. 1971Stanford prison simulationA six-day study became a global story about roles overpowering character.Historical influence: the study shaped textbooks, film and institutional debate. Current status: coaching, selection, researcher participation and weak controls severely limit the classic conclusion. 1975–99Moral disengagement frameworkBandura organized the cognitive mechanisms that permit harm without abandoning a moral self-image.Contribution: the model explains how language, responsibility and victim representation transform self-judgment. It is not a checklist proving intent in a specific criminal case. 2002–06BBC Prison StudyParticipants did not simply adopt roles; organization, identification and resistance changed over time.Contribution: tyranny requires shared identity, leadership and a workable project. Groups can also organize against arbitrary power. 2004–09Abu Ghraib and institutional accountabilityImages of prisoner abuse revived arguments about bad actors versus command climate.Lesson: individual criminal responsibility and institutional causation can coexist. Training, leadership, rules, resources and reporting systems shape what conduct becomes possible and punishable. 2018–19Archival audit of StanfordNew analyses challenged the most famous details of the study’s origin story.Correction: the record supports a study influenced by expectations and leadership rather than spontaneous universal corruption. The lesson becomes one about guided enactment, not automatic role possession. CurrentMulti-level preventionViolence research increasingly combines individual, relational, organizational and structural interventions.Current direction: prevention focuses on reducing opportunity, strengthening accountability, interrupting escalation, supporting victims and changing conditions that normalize harm. Incremental commitment can narrow perceived alternatives one step at a time. Photo: Valentin Lacoste / Unsplash 05Definitions Before DiagnosisWords that are related but not interchangeable Moral categoryEvilA judgment that an act or actor represents profound wrongdoing. It communicates condemnation but does not specify mechanism, diagnosis, intent or legal element. Behavioural categoryAggressionBehaviour intended to harm. Aggression can be physical, verbal, relational, impulsive, defensive, instrumental or institutionally authorized. Public-health categoryViolenceThe intentional use of physical force or power, threatened or actual, against oneself, another person, a group or community, with likely injury or deprivation. Experiential categoryCrueltyInflicting, exploiting or disregarding suffering. Cruelty may involve pleasure, indifference, punishment, humiliation or instrumental calculation. Legal categoryCulpabilityResponsibility under a defined law. It depends on conduct, mental state, defences, evidence and procedure, not a general personality judgment. Analytical categoryMalevolenceA pattern of wishing, planning or accepting serious harm. It can describe intent, but it is not a standardized clinical diagnosis. Traits and biological signals are evidence only within validated limits, not verdicts. Photo: Arturo Castaneyra / Unsplash 06Moral DisengagementEight mechanisms that can make prohibited conduct feel permissible Moral justificationHarm is reframed as serving security, justice, purity, loyalty, defence or a higher cause. The actor does not abandon morality; the actor recruits morality to authorize the act. Euphemistic labellingLanguage sanitizes consequences. Killing becomes “neutralization,” abuse becomes “enhanced procedure,” and civilian suffering becomes “collateral.” Vocabulary creates distance between decision and injury. Advantageous comparisonA harmful act is made to appear minor by comparing it with something worse. The relevant question, whether the act itself was justified, is displaced by a more extreme example. Displacement of responsibilityThe actor locates agency in a superior, law, policy or chain of command. “I followed orders” describes pressure but does not automatically eliminate moral or legal responsibility. Diffusion of responsibilityTasks are divided so that no participant experiences the whole act as theirs. Committees, teams and bureaucracies can distribute action while preserving the harmful outcome. Disregarding consequencesDistance, secrecy, metrics and abstraction make injury less visible. A person who never sees the victim can still make the decision that causes the harm. DehumanizationVictims are described as vermin, disease, cargo, threats, numbers or inferior beings. Once humanity is denied, ordinary reciprocity becomes easier to suspend. Attribution of blameThe victim is said to have provoked, deserved or forced the conduct. Responsibility is transferred from the decision-maker to the person harmed. Responsibility can be displaced or diffused, but action is still carried out by people. Photo: Marco Chilese / Unsplash 07Obedience Under AuthorityWhat Milgram showed, and what the experiment cannot settle Supported lessonEscalation changes the decisionMilgram did not ask participants to begin with the maximum switch. Small increments made each next action resemble the previous one. Commitment, laboratory legitimacy and the experimenter’s prompts created pressure to continue.Variation studies matter as much as the headline result. Compliance changed when the authority was less proximate, peers resisted, the setting changed or participants had to act more directly. CorrectionCompliance was not automaticMany participants challenged the experimenter, displayed distress or stopped. Later interpretations argue that some continued because they identified with the scientific project, trusted the setting or believed the learner was not in real danger.The evidence supports vulnerability to legitimate authority, not a claim that conscience disappears whenever a person in a coat issues an order. Confinement concentrates authority, dependence and vulnerability. Photo: Fabio Sasso / Unsplash 08Stanford Prison Experiment AuditA famous narrative with serious evidentiary defects MythRandomly assigned roles caused ordinary guards to become uniformly sadistic.RecordGuard conduct varied substantially. Archival work indicates that expectations, instructions and researcher leadership helped shape the aggressive role. MythThe study functioned like a controlled experiment proving the power of situation.RecordThere was no conventional control group, researchers were embedded in the simulation, variables were not isolated and the sample was small and self-selected. MythPrisoners and guards simply forgot they were participants.RecordParticipants knew they were in a study, interpreted expectations and sometimes performed strategically. Their awareness does not make the distress unreal, but it changes causal interpretation. MythThe experiment proves institutions are more powerful than individual responsibility.RecordThe better lesson is that leaders, norms and institutional design influence behaviour while participants retain different capacities to comply, resist, organize or withdraw. Architecture can organize surveillance, obedience and unequal control. Photo: Nik Owens / Unsplash 09Group Identity and LeadershipTyranny requires a shared project, not merely a crowd Identification“We” determines what becomes normalPeople do not lose all identity in groups. They often shift from personal identity to a shared identity. Conduct then follows the group’s norms: rescue in one group, persecution in another.A leader can define who belongs, what threat exists, which conduct proves loyalty and whose suffering should be ignored.Older deindividuation research linked anonymity and emotional arousal to weaker self-restraint in crowds. The same research also found generosity, cooperation and collective joy under those conditions. The process is neutral. Group norms, leadership and what the setting rewards decide whether it ends in solidarity or a mob. ResistanceGroups can also protectShared identity can create whistleblowing, solidarity, mutual defence and refusal. The same capacity for coordination that enables abuse can organize resistance when the group’s moral norm rejects harm.Prevention therefore requires more than telling individuals to “be brave.” Institutions must make dissent legitimate, visible and materially safe. Shared identity can mobilize persecution, solidarity or resistance. Photo: Amir Arabshahi / Unsplash 10Dehumanization and Moral ExclusionHow language narrows the circle of obligation Animalistic“Vermin” and “beasts”Targets are described as lacking refinement, reason or restraint. Violence is then framed as containment, cleansing or pest control. MechanicalNumbers and objectsPeople become units, inventory, files, targets or throughput. Suffering disappears behind technical language and administrative process. Moral exclusionOutside the boundaryFairness and compassion are treated as obligations owed only to the in-group. The victim is not always denied humanity outright; they may be declared undeserving of protection. Victim erasure warningDo not turn harm into an offender-centred thought experimentDiscussions of “ordinary people doing evil” can make victims disappear behind a fascinating mechanism. Every laboratory analogy, military report or institutional case ultimately concerns a person who experienced fear, injury, humiliation, loss or death. The mechanism matters because the suffering was real.Dehumanization converts people into categories, contaminants or obstacles. Photo: Cem Salini / Unsplash 11Bystanders, Diffusion and SilenceInaction is shaped by uncertainty, coordination and perceived permission AmbiguityPeople hesitate when they are unsure whether an event is dangerous, private or already controlled. Institutions can reduce ambiguity through clear thresholds and reporting rules. Pluralistic ignoranceEach observer uses other observers’ calm behaviour as evidence that intervention is unnecessary. Everyone may be privately concerned while the group appears unconcerned. Diffusion of responsibilityThe presence of many witnesses can reduce the felt obligation of each individual. Naming a specific person and action can restore responsibility. Retaliation riskSilence may be rational when reporting threatens employment, status, safety or immigration security. Ethical analysis must examine the retaliation system, not only the courage of the witness. Institutional betrayalOrganizations can invite reports and then punish, expose or discredit the reporter. Failed response teaches the wider group that silence is safer than disclosure. Clear invitations and specific responsibility can turn bystanders into responders. Photo: Edward Kucherenko / Unsplash 12Victimology Without StigmaWho is harmed, who is believed and whose suffering becomes invisible Exposure patternPower shapes victimizationPeople with less institutional power may face greater exposure to abuse and fewer safe exit routes: prisoners, children, employees, patients, migrants, people experiencing homelessness and members of stigmatized groups.“Vulnerability” describes an unequal environment. It is not a flaw in the person harmed. Mental-health correctionDiagnosis is not a violence labelMost people with mental illness are not violent. Risk is better understood through combinations such as prior violence, acute crisis, substance misuse, access to weapons, victimization and social instability.People with serious mental illness also experience high rates of victimization. Reporting that presents diagnosis as the central explanation can increase stigma while doing little to improve prevention. Protective relationships and credible support can interrupt pathways to harm. Photo: Mita 64 / Unsplash 13Predisposition Without DestinyBiology can influence risk; it cannot supply a verdict NeurobiologyRegulation and threatResearch examines networks involved in impulse control, reward, threat detection, empathy and decision-making. Group averages cannot diagnose a violent individual.The most cited pattern pairs weaker prefrontal inhibition with an overactive amygdala threat response, and childhood maltreatment, trauma and substance use can amplify it. GeneticsInteraction, not a “violence gene”Genetic variants may affect sensitivity to environments or regulation. Effects are small, distributed and dependent on development and context.The best-known example, a low-activity variant of the MAOA gene, is linked to aggression mainly in people exposed to stressful or adverse environments. PersonalityTraits are not crimesCallousness, antagonism, impulsivity and sensation-seeking can be relevant to risk formulation, but traits neither prove future violence nor replace evidence of conduct. SubstancesAcute disinhibitionAlcohol and some drugs can increase impulsivity, misperception and conflict. The effect depends on dose, expectations, setting, history and access to victims or weapons. Past conductBehaviour predicts better than labelsDocumented prior violence, coercive control, threats and escalating boundary violations are generally more informative than broad demographic or diagnostic stereotypes. Forecast limitRare events defeat certaintyEven useful risk tools generate false positives when the outcome is rare. Responsible assessment communicates uncertainty and focuses on changeable conditions. 14Trauma, Adversity and Learned ViolenceExposure can raise risk without predetermining replication DevelopmentWhat adversity can changeChronic maltreatment, neglect, household violence and unstable caregiving can affect stress response, trust, emotion regulation and beliefs about power. Children may learn that aggression is effective or unavoidable.Those pathways increase need for support; they do not make later offending inevitable. Protective factorsWhat interrupts transmissionSafe relationships, stable housing, treatment, education, meaningful belonging, economic security and nonviolent models can alter developmental trajectories.The phrase “cycle of violence” should describe a preventable risk process, not a prophecy imposed on survivors. 15Systems That Enable CrueltyHow organizations convert individual choices into durable patterns IncentivesReward the outcome, ignore the methodTargets, quotas, promotions and informal praise can reward harmful results while policies officially prohibit the means used to obtain them. SecrecyControl the informationRestricted access, classification, nondisclosure, fragmented records and intimidation can prevent outsiders from seeing a pattern that insiders experience as routine. ImpunityPredictable non-enforcementA rule has little protective value when powerful offenders expect delayed, selective or symbolic consequences. DistanceSeparate decision from sufferingAutomation, subcontracting and bureaucracy can place the person making a harmful decision far from the person absorbing it. NormalizationYesterday’s breach becomes today’s routineSmall tolerated violations create precedents. New members learn the real rule from what is rewarded and ignored, not from the written code. ScapegoatingContain the scandalOrganizations may isolate responsibility in one visible “bad apple” to protect the leadership, design and incentives that made recurrence likely. Secrecy and restricted visibility make institutional harm harder to challenge. Photo: Samuel Jerónimo / Unsplash 16Evidence ToggleWhat different research traditions can and cannot establish Experiments Field Evidence Clinical Risk Systems Review Laboratory evidenceControlled pressure with limited realismExperiments can isolate prompts, proximity, group behaviour and incremental commitment. They provide causal leverage but may rely on unusual samples, artificial tasks and participants who infer the researcher’s expectations. Laboratory compliance is not identical to criminal violence.Causal leverageDemand effectsExternal validity 17Media MythologyWhy simple stories outlive complicated evidence MythThere are “evil brains” that can be recognized before the act.RecordNeuroscience identifies group-level associations and impaired processes. It does not provide a reliable scan-based test for future cruelty. MythTrauma creates killers.RecordTrauma is common; severe violence is rare. Most survivors do not offend, and many violent actors do not share one trauma profile. MythStanford proved anyone becomes sadistic in uniform.RecordThe study was methodologically compromised. Later work points to leadership, identification and expectations, not automatic role capture. MythObedience removes responsibility.RecordAuthority can pressure and shape decisions. Legal and moral responsibility still depends on capacity, knowledge, alternatives and conduct. MythOnly monsters commit atrocities.RecordSome perpetrators show persistent predatory traits; others act through ideology, loyalty, fear, careerism or normalized institutional practice. One model cannot explain every case. Public narratives simplify evidence; collective dissent can restore suppressed moral boundaries. Photo: Lauren Mitchell / Unsplash 18Legal ResponsibilityExplanation, mitigation, justification and excuse are different questions Intent and knowledgeCriminal law often distinguishes purposeful, knowing, reckless and negligent states of mind. A psychological explanation does not automatically alter the state the prosecution must prove. Duress and superior ordersThreats and command structures may be relevant, but the legal effect varies by jurisdiction and offence. “I was ordered” is not a universal defence. Mental disorderA diagnosis is not the same as legal incapacity. Courts ask specific questions about understanding, appreciation, control or participation in proceedings. MitigationTrauma, coercion, youth, impairment and social context may affect sentence or culpability without declaring the act justified or harmless. Institutional accountabilityFocusing only on the direct actor can miss command decisions, negligent supervision, evidence destruction, retaliation and policies that produced foreseeable abuse. Legal responsibility depends on conduct, capacity, evidence and process. Photo: Jan Oblak / Unsplash 19Prevention ArchitectureDesign conditions in which resistance is easier and abuse is harder IndividualRecognize escalationTeach people to identify coercive scripts, euphemism, victim blaming, incremental commitment and dehumanizing language before the final harmful act. RelationshipStrengthen safe challengeBuild peer cultures where dissent is respected, coercive control is recognized and help-seeking does not carry automatic loss of status. OrganizationSeparate power and reviewUse independent oversight, auditable decisions, rotation in high-discretion roles, protected reporting and meaningful consequences. EnvironmentReduce opportunityImprove supervision, lighting, staffing, access controls, record integrity and victim support without converting prevention into indiscriminate surveillance. Public healthAddress modifiable risksStable housing, violence-interruption programs, substance-use treatment, mental-health care and early family support can reduce multiple forms of harm. CivicProtect truth-tellersWhistleblower protection, investigative journalism, transparent records and independent courts make it harder for powerful systems to define abuse as normal. 20Unresolved VoidQuestions the evidence does not allow us to close How much does ideology matter?Are people primarily pressured by authority, or do they comply because the authority represents a cause they already value? Different cases likely require different answers. When does gradual escalation become deliberate strategy?Small steps can trap a participant, but sophisticated offenders and institutions may deliberately design those steps to recruit complicity. Can empathy be selective and still restrain harm?People may show deep care for an in-group while supporting violence against an out-group. General empathy scores can miss that boundary. How should rare violence be predicted?Low base rates create false positives. A system may identify many “high-risk” people while being wrong about most of them. When does prevention become coercion?Surveillance and preventive detention can reproduce the same dehumanization they claim to stop. Proportionality, review and due process remain essential. How do institutions learn after scandal?Public condemnation can produce new policies without changing incentives, leadership or retaliation. The meaningful measure is recurrence, not announcement. 21Sources and VerificationPrimary research, methodological critique, ethics and public-health contextEditorial stock photography is used illustratively. Each image is individually credited and linked to its Unsplash source; no stock image is reused on this page. Primary experimentMilgram: Behavioral Study of ObedienceThe 1963 paper reports the best-known baseline obedience condition and provides the original procedure and observed participant behaviour.Open DOI record Core frameworkBandura: Moral DisengagementThe 1999 review describes mechanisms that disconnect moral self-sanction from harmful action.Open DOI record Group identityReicher and Haslam: BBC Prison StudyThe study challenges automatic role internalization and examines identification, power, organization and resistance.Open DOI record Method auditLe Texier: Debunking StanfordAn archival analysis of the Stanford Prison Experiment’s design, instructions, reporting and enduring textbook mythology.Open DOI record Research ethicsThe Belmont ReportFoundational principles for human-subject research: respect for persons, beneficence and justice.Open HHS record Public-health frameworkWHO World Report on Violence and HealthA multi-level framework for understanding violence, risk, prevention and institutional response.Open WHO report Case file · FAQ Frequently asked questions Is cruelty a single personality trait? No. Cruelty is not a single trait. It is a process of agency, identity, authority, opportunity and permission. What is moral disengagement? Cognitive and social processes that suspend self-condemnation, allowing a person to participate in harm while preserving a favourable self-image. Bandura’s framework identifies eight recurring ways people neutralize self-sanction. Did the obedience experiments show universal obedience? No. In the best-known baseline condition, 26 of 40 participants continued to the final switch. The result demonstrates pressure, not mindless universal obedience. Can science predict who will commit violence? No brain scan, gene, diagnosis or childhood event can identify an inevitable violent offender. Risk factors are probabilistic and context-dependent. What is dehumanization? Treating people as less than fully human, or as objects, contaminants, numbers, enemies or obstacles.