Banality of evil and the problem of many hands

Working Definition

The banality of evil and the problem of many hands are two complementary concepts explaining how large organizations, bureaucracies, and technical systems produce severe harm without the involvement of a single demonic villain. The banality of evil describes the psychological and cognitive dimension: the capacity of an ordinary person to participate in a destructive enterprise by focusing unreflectively on procedural rules, officialese, and personal career advancement while switching off moral imagination. The problem of many hands describes the structural dimension: a situation in which a final decision or disaster is the composite result of small, partial contributions from so many specialists, committees, and technical modules that no single participant satisfies the classical requirements of full individual culpability (simultaneous intent, knowledge, and direct causal control). Taken together, they show how the modern division of labor can produce a crime or catastrophe without an identifiable perpetrator.

Genesis and Global Lineage

German-American political philosopher Hannah Arendt introduced the phrase “banality of evil” in her 1963 book Eichmann in Jerusalem (an expanded version of her five-part report for The New Yorker on the 1961 trial of Adolf Eichmann), drawing heavily on Raul Hilberg's pioneering monograph The Destruction of the European Jews (1961) on the administrative machinery of the Holocaust. Arendt broke with the romantic-theological tradition of portraying radical evil as the work of satanic sadists, pointing instead to the intellectual shallowness of a bureaucrat who thought exclusively in stock phrases and departmental circulars. Seventeen years later, working in the very different register of analytical political philosophy, American political scientist Dennis F. Thompson published Moral Responsibility of Public Officials: The Problem of Many Hands in the American Political Science Review (1980). Analyzing U.S. government decision-making, Thompson demonstrated that both the strictly hierarchical model (only the person at the top is responsible) and the collective model (the entire institution is responsible as a legal person) serve in practice as shields of impunity. In 1996, philosopher of technology Helen Nissenbaum transferred Thompson's concept to computer systems, opening the contemporary debate on diffused accountability in software engineering and artificial intelligence.

Central European Lineage

Both the historical material underlying Arendt's diagnosis and its deepest sociological elaboration are closely tied to Central Europe. The primary subject of Eichmann's trial was the logistics of deporting Jews from Vienna, Prague, and occupied Poland, culminating in the spring and summer of 1944 with the deportation—carried out with the collaboration of the Hungarian administration—of more than four hundred thousand Hungarian Jews to Auschwitz-Birkenau. Yet even before Arendt formulated her thesis in English, the mundane, bookkeeping-and-industrial character of the camp division of roles had been captured in postwar Poland by writer Zofia Nałkowska in her documentary cycle Medaliony (Medallions, 1946) and by Auschwitz survivor Tadeusz Borowski in his stories from 1946–1948, which depicted the camp as a rationally managed enterprise where mass murder dissolved into the routine of ramps, warehouses, and card indexes. In 1989, this line of analysis was taken up by sociologist Zygmunt Bauman—educated and long active at the University of Warsaw before being purged in the antisemitic campaign of March 1968—in his landmark book Modernity and the Holocaust. Bauman introduced the concept of “adiaphorization”: the mechanism whereby the bureaucratic division of tasks, mediation through a chain of executors, and the substitution of technical responsibility for moral responsibility remove human actions from the judgment of conscience, rendering them morally neutral in the eyes of the performer.

Why Now

Between 2018 and 2026, the pairing of the banality of evil and the problem of many hands became a primary framework for analyzing institutional and algorithmic failures. First, major scandals surrounding automated welfare states—such as the Dutch childcare benefits scandal (Toeslagenaffaire, which brought down Mark Rutte's government in 2021 after an algorithmic risk-scoring system and tax officials falsely accused tens of thousands of families of fraud) and Australia's Robodebt scheme (examined by a 2023 Royal Commission)—displayed “digital adiaphorization” in pure form: caseworkers rejected appeals from desperate citizens by pointing to a computer screen, while programmers pointed to ministry guidelines. Second, in civil aviation, the Boeing 737 MAX crashes of 2018–2019 (investigated in the 2020 U.S. House Transportation Committee report) exposed how splitting safety certification across subcontractors, MCAS engineers, managers cutting pilot training costs, and the FAA regulator meant that no single actor felt like the author of a lethal trap. Third, the deployment of generative AI value chains (2023–2026) and negotiations over the EU AI Act revealed an extreme problem of many hands: when an autonomous agent or recommender system causes harm, responsibility scatters across the training-data provider, the foundation-model developer, the fine-tuning firm, the API integrator, and the institutional end user.

Explanatory Power

These concepts break the naive legal and everyday intuition that, in the face of any massive harm, looks for a “guilty mind” (mens rea) proportional to the magnitude of the outcome. In complex socio-technical systems, the scale of a catastrophe is completely decoupled from the malice of any single participant. An engineer's minor compromise in closing a bug ticket before a sprint deadline; a product manager's decision to optimize user retention by half a percent; a clerk's habit of never looking beyond a checklist; and a machine-learning model's minimization of a loss function—each of these steps taken in isolation falls within ordinary professional norms. Only when coupled into a single decision pipeline do they form a ruthless machine that grinds down human lives. The framework of Arendt, Thompson, and Bauman also explains why parliamentary inquiries and investigative commissions so often encounter genuine feelings of innocence among all those questioned after a scandal breaks: each person truly saw only their narrow slice of the interface and judged their own work solely by the standard of technical diligence.

Axes of Debate

Underappreciated Angle

In the twenty-first century, the problem of many hands has ceased to be merely an unintended byproduct of growing complexity—it has become a consciously designed market commodity that might be called “diffused responsibility as a service.” When a corporate board or government ministry wants to carry out drastic layoffs, tighten credit-scoring rejections, or cut supplier rates, the hardest barrier is not a lack of data but human moral resistance and reputational risk. Hiring an external consulting firm that deploys an “objective predictive model” supplied by a third-party vendor and hosted on a fourth provider's cloud deliberately multiplies the number of hands in the causal chain. The more layers of abstraction, API interfaces, and external audits separate the person signing the contract from the human being losing a job or welfare benefit, the easier it is for every link in the chain to keep a clear conscience. In this way, the architecture of microservices and subcontracting becomes a modern machine for laundering moral responsibility.