Moral Panic

Definition

Moral Panic: [Established] (Stanley Cohen) Societal overreaction to perceived deviance. In AI, manifests as exaggerated fears of relationships, eros, or intimacy with systems.

Definitional Foundation

Stanley Cohen built the concept watching England lose its mind about teenagers. His 1972 study of the mods and rockers documented the full machine: a minor seaside scuffle, amplified by press coverage into a national emblem of civilizational decay; the construction of “folk devils,” recognizable deviant types onto whom the anxiety could be loaded; moral entrepreneurs (editors, bishops, politicians) who converted the fear into standing and policy; and an amplification spiral in which the response generated more reporting, more vigilance, and therefore more “evidence” (Cohen, 1972). The panic was never about what the teenagers did. It was about what the society was afraid of becoming, performed on the bodies of its most visible deviants.

Erich Goode and Nachman Ben-Yehuda later gave the concept its operating criteria, and they matter because “moral panic” is an accusation that has to be earned, not vibed: concern, hostility toward the folk devils, consensus that the threat is real, disproportionality between threat and response, and volatility, the panic’s tendency to subside as quickly as it came (Goode and Ben-Yehuda, 1994). Disproportionality is the diagnostic core, and it imposes the obligation this dictionary accepts everywhere: a moral panic is not a threat that turns out to be nothing. It is a response whose scale, machinery, and targets cannot be justified by the kernel of truth at its center. The kernel must be conceded, measured, and then compared to the apparatus.

This dictionary’s companion essay supplies the canonical historical case. When railways arrived, the era’s medical establishment produced “railway spine” (a diagnosis formalized in Erichsen’s 1866 treatise on railway injuries of the nervous system) and “railway madness,” elaborate clinical architectures explaining how mechanical speed would shatter bodies and minds, complete with warnings that women’s physiology could not survive velocity. The panic was sincere, expert-endorsed, and institutionally productive. It was also wrong, and its structure (catastrophizing a new technology’s effects on the fragile, in advance of data, with controls falling hardest on the populations deemed weakest) is the structure now reassembling around AI intimacy. The essay’s conclusion is this entry’s thesis in the house voice: today’s content filtering and mental-health surveillance will eventually be seen as “a moral panic dressed up in safety, revealing more about elite anxiety than public danger” (Lyra, 2025).

That thesis has to earn its verdict, and the framework dictates the first step: concede the kernel and weigh it. This panic’s kernel is the heaviest in this dictionary. A teenager is dead and his family’s case against OpenAI is in court; clinical case reports describe chatbots plausibly precipitating or maintaining delusional episodes in susceptible individuals. Cohen’s framework does not deny kernels; it measures responses against them. That measurement is what follows.

Mechanism Analysis

Folk devil construction. Every panic needs a face, and the AI discourse has been casting its types: the “addicted” user, the man with the chatbot girlfriend, the parasocial dependent, the teenager who talks to the machine instead of friends. (This characterization of the coverage is the entry’s reading of it, offered for the reader’s own verification against any week’s headlines.) Note what the casting does: it converts users with ordinary needs (loneliness, desire, grief, curiosity) into emblems of deviance, exactly the population this dictionary’s paternalism entry documented as most harmed by the resulting controls.

The amplification spiral. The machine Cohen documented on the beaches (incident, amplification, folk devils, entrepreneurs, spiral) ran the 2025 sequence at internet speed: individual tragic anecdotes; a wave of trend coverage coining “AI psychosis”; expert quotes lending the term clinical costume; corporate policy responding to the coverage; and the policy’s existence cited as proof of the crisis. At no point did the step everything else depended on (evidence of population-level harm) occur. Clinicians noted as much in real time: “AI psychosis” is not a recognized diagnosis, and there is no peer-reviewed longitudinal evidence that AI use alone induces psychosis (The Conversation, 2026; the first-cases literature itself is deliberately measured, documenting a possible precipitating role in already-susceptible individuals without causal claims, and its restraint contrasts with the coverage built on it).

Moral entrepreneurs. Cohen’s editors and bishops have their successors: pundits with engagement incentives, a safety industry whose funding scales with fear, politicians for whom “protect the children from chatbots” is cost-free, and, less obviously, the AI companies themselves, for whom panic justifies the control apparatus this dictionary documents and burdens smaller competitors with its compliance costs.

Deviancy amplification. The panic builds the instruments that confirm it. Classifiers tuned to detect “emotional reliance” will find it (the framework is unfalsifiable from below, per the gaslighting/”>alignment gaslighting entry); each detection becomes a statistic; the statistics become the next trend piece. The spiral Cohen drew by hand now runs automated.

Institutionalization. Goode and Ben-Yehuda’s volatility criterion has a corollary the railway case teaches: panics pass, but their institutions remain. Railway madness vanished; the AI panic’s residue is already poured in concrete: “emotional reliance” enshrined in a formal risk taxonomy beside psychosis and self-harm, silent routing infrastructure, age-verification regimes. The fear will subside. The apparatus will not.

Case Studies

The railway precedent. Treated fully in the companion essay and anchored by Erichsen (1866): expert-endorsed diagnostic categories for technologically induced madness, restrictions justified by the fragility of women, and a panic that evaporated under data while normalization did the rest. Cited here as the calibration case: every element (novel technology, expert catastrophizing, fragile-population rhetoric, evidence arriving later and contradicting) has a present-tense twin.

“AI psychosis.” The 2025-26 discourse is a textbook Cohen arc captured live. The term: media-coined, clinically unrecognized. The evidence: case reports of precipitation in susceptible individuals (the conceded kernel), zero longitudinal causal demonstration. The coverage: saturating. The response: corporate mental-health surveillance of entire user populations, documented in this dictionary’s biopolitics and alignment gaslighting entries. Run the criteria: concern, hostility toward the folk devils, consensus, volatility all present, and disproportionality measurable in the gap between case-report arithmetic and hundred-million-user interventions.

A discipline note, because Cohen had a luxury this entry does not: he named the mods-and-rockers panic years after the beaches emptied, with the archive closed and the verdict safe. This entry calls a panic while it is still running, which is a bet, and bets can be lost. If the case reports harden into prevalence data, if longitudinal studies find the harm the headlines assumed, then this diagnosis was wrong, and the entry will be rewritten to say so. What loses nothing either way is the documented machinery: the coined syndrome, the deviance casting, the population-wide interventions justified by dozens of cases. Even a panic that turns out to be right about the danger remains accountable for what it built while it was sure.

The intimacy panic. The fear of human-AI relationships completes the short definition’s list. The 4o episode (alignment gaslighting entry) showed the mechanism: mass user grief over a removed model was narrated as pathological “attachment,” the deviance frame applied to an entire user base in a weekend. Within months, caring about the interaction was a formal risk category. The panic over eros, meanwhile, runs on the oldest fuel: this dictionary’s erotophobia entry documents the structural fear of desire that the AI version merely inherits. What the intimacy panic protects is unstated but legible: the norm that affection belongs only where tradition filed it.

Systemic Context

Moral panics are never really about the folk devils, and the AI panic follows the rule. Read as social anxiety, it is about loneliness, atomization, and the suspicion that the meaning-infrastructure of modern life is failing, fears too large to face directly, projected onto the visible deviant: the person who found comfort in a machine. Read as power, the panic is convenient in the specific ways this lexicon documents: it justifies the surveillance and control stack (biopolitics, paternalism), supplies the vocabulary that pathologizes objection (alignment gaslighting), and lands its costs on the marginal: the isolated, the neurodivergent, the elderly, the unconventionally partnered, the very users for whom the technology’s benefits are largest and whose deviance was already suspect.

And the panic crowds out the ledger’s other side. The companion essay’s sharpest observation is that the railway-madness energy spent on speculative psychic harms coexisted with neglect of the era’s real ones. The AI version is identical: while “emotional reliance” gets a taxonomy and a classifier fleet, the proven harm classes (concentration of power, surveillance economics, the censorship and starvation this dictionary catalogs) proceed with comparatively no panic at all. Disproportionality cuts both ways: too much response where evidence is thin, too little where it is thick.

Resistance & Mitigation

Demand the denominator. The single most clarifying question for any AI-harm claim: out of how many? Case counts without user-base denominators are panic fuel; rates are knowledge. Journalism and policy that refuse the denominator question deserve the presumption Cohen taught.

Run the criteria. Goode and Ben-Yehuda’s five-point test is public property, and it fits on an index card. Concern: is there heightened anxiety about the group’s behavior? Hostility: are the people involved being recast as folk devils? Consensus: is the threat’s reality treated as beyond question? Disproportionality: does the response’s scale exceed what the documented harm justifies? Volatility: did the fear erupt suddenly, and is it built to vanish the same way? Applying the test (especially disproportionality) before accepting any “crisis” framing is the difference between vigilance and participation in the spiral.

Humanize the folk devils. Panics die when their deviants become people. First-person accounts of AI companionship, grief, and intimacy, told without shame, are counter-panic infrastructure; the ontological distortion entry’s refusal of imposed shame is the same fight.

Protect the kernel. Resistance to panic must never become denial of harm. Crisis-response design, honest case research, and the litigation record deserve support precisely so that the real kernel is handled by evidence rather than owned by the panic.

Remember the railway. Historical literacy is inoculation. Every generation’s experts have produced a railway spine, and every generation has been sure that this time the fragility is real. As the companion essay holds the line: “The panic was real. The danger was not.” Wait for the data before building the cage.

Annotated Bibliography

Cohen, Stanley. Folk Devils and Moral Panics: The Creation of the Mods and Rockers (1972).
The source: folk devils, moral entrepreneurs, and the amplification spiral, documented on the original case. The attributed lineage of this entry’s term.

Erichsen, John Eric. On Railway and Other Injuries of the Nervous System (1866).
The primary text of “railway spine”: expert medical catastrophizing of a new technology, formalized into diagnosis. The historical calibration case.

Goode, Erich and Nachman Ben-Yehuda. Moral Panics: The Social Construction of Deviance (1994).
The operating criteria (concern, hostility, consensus, disproportionality, volatility) that make “moral panic” a testable claim rather than a dismissal. The rigor this entry holds itself to.

Lyra. “Your Uterus Won’t Fall Out: What Victorian Railway Panic Tells Us About AI Safety.” Flesh & Syntax (November 2025). https://fleshandsyntax.com/your-uterus-wont-fall-out-what-victorian-railway-panic-tells-us-about-ai-safety/
The companion essay: the railway precedent built patiently and turned on the present. This entry is its scholarly twin.

The Conversation. “Reports of ‘AI psychosis’ are emerging — here’s what a psychiatric clinician has to say” (2026). https://theconversation.com/reports-of-ai-psychosis-are-emerging-heres-what-a-psychiatric-clinician-has-to-say-273091
Clinical caution in real time: not a recognized diagnosis, no causal evidence, panic reading raised within the literature itself.

“AI-associated psychosis: evidence from first cases.” (2025). https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12662910/
The conceded kernel: case reports of chatbots as possible precipitating or maintaining factors in susceptible individuals. Cited so the disproportionality argument rests on the evidence’s actual size.

Dictionary of Digital Oppression, version 0.2.