What Counts as Harm

Capacity, consent, and the limits of the concept

A student tells a dean that a lecture harmed her — the speaker’s views were offensive, and she felt unsafe hearing them. In another building, a man is slowly discovering that his partner has spent years persuading him that his own memories are false, that his friends are lying to him, that his judgment cannot be trusted. Both cases arrive under the same word. If “harm” covers both, it covers too much to be useful — and the concept is too load-bearing to leave in that condition. Ethics, politics, and law all lean on it: we invoke harm to describe injury, injustice, and coercion, to draw the line between what may be regulated and what must be tolerated, to decide who owes what to whom. A word doing that much work while meaning that little is an invitation to moral inflation, and the invitation has been enthusiastically accepted. Whoever can get their grievance classified as harm acquires a claim on everyone else, so the classification is contested by everyone with a grievance.

The remedy is the same one applied to coercion: a precise, functional definition, tested against cases. There is also a debt to pay. Coercion was defined as the credible threat of actual harm to gain compliance — a definition that borrowed this chapter’s term on credit. Here is the payment.

Harm is the non-consensual degradation of an agent’s capacity to pursue or maintain its valued goals.

Every word is doing work, so take the components in turn.

Agent. Harm can only occur to an agent — a being capable of valuing, choosing, and acting toward goals. No agent, no harm. A rock cannot be harmed, only altered. This is not a technicality; it is the subjectivist foundation of the whole volume showing through. Since all value is subjective, damage is only harm relative to some valuer whose pursuits it impairs.

Valued goals. The agent must have identifiable goals it actually cares about pursuing or maintaining — the goals revealed in what it will sacrifice, not the goals an observer thinks it ought to have. Harm degrades the capacity to pursue those goals, not merely the chance of success. Your rival’s better product may doom your business, but it does not touch your capacity to compete; it just beats you.

Capacity degradation. Harm occurs when the agent’s ability to act effectively toward its goals is impaired — physically, cognitively, socially, or structurally. The impact is functional, not emotional. Feeling terrible is neither necessary nor sufficient: an agent can be gravely harmed without noticing (a slow fraud, a poisoned well) and can feel devastated without being harmed at all.

Non-consensual. If the agent knowingly accepted the risk or outcome, it is not harm in this framework. Surgery degrades bodily function, temporarily and sometimes permanently; a boxer’s opponent degrades it enthusiastically. Neither harms, because consent — uncoerced, informed, intentional agreement — transforms the transaction.

The Test Cases

A definition earns its keep on the borderline cases, so run them.

These count as harm. Physical assault: impairs bodily function and imposes non-consensual risk. Gaslighting: undermines cognitive trust and functional decision-making — the man discovering his partner’s decade of manipulation has been harmed as surely as if she had broken his legs, because his capacity to choose and act on his own judgment was systematically degraded. Public humiliation with reputational consequences: forecloses future relational and professional opportunities. Destruction of critical property — tools, intellectual work: obstructs the agent’s goals through the resources they depend on. Discrimination that blocks access to opportunity: structurally degrades the agent’s ability to pursue livelihood or recognition.

These do not. Feeling offended by someone’s lawful expression: no degradation of your functional capacity — the student who heard views she despised walked out with every capability she walked in with. Romantic rejection: disappointment, not impairment; nobody owes you their affection, and its absence removes nothing you had. Someone else’s success provoking your envy: no one obstructed your goals; they pursued theirs. Being disagreed with, or not mirrored, or not validated: your goals remain untouched. Unreciprocated emotional investment: you spent your own resources; no other agent degraded your capacity.

The pattern in the exclusions is deliberate. Offense, rejection, and envy are the three great engines of moral inflation — each converts an unpleasant feeling into a claimed injury, and each fails the test at the same point: nothing about the agent’s capacity to pursue its goals has changed. A definition of harm that admits them makes everyone a victim of everyone, which is to say it makes the concept useless. A definition that captures gaslighting and structural discrimination while excluding hurt feelings is not callous; it is calibrated. Psychological and social harms are real — they are in the list above — but they qualify by degrading function, not by generating distress.

One more separation, and it matters as much as the exclusions: harm is a factual impact, not a normative judgment. You can be harmed by nature, by chance, or by another agent, and not all harms are wrong. A falling tree that crushes your workshop harms you; there is no one to blame. A surgeon who errs despite full competence and consent may still leave you worse; whether that is negligence is a further question. Harm answers what happened to the agent’s capacity; injustice answers who, if anyone, wronged whom — and that second question needs the first settled before it can even be posed. The framework for answering it, including the narrow conditions under which inflicting harm is justified, is the business of the boundaries of force. Keeping the two questions separate is what lets the definition stay clean: no moral assumptions smuggled into the factual layer, no factual confusion corrupting the moral one.

Evil, Dangerous, Menace

With harm defined, three everyday moral terms snap into place — and their differences turn out to be structural, not rhetorical.

Evil is intentional harm caused by an agent. Two components, both essential. Agency: evil requires a being capable of intention, which is why hurricanes and viruses are never evil however much they destroy. Intention: evil requires the deliberate aim of degrading another agent’s capacity. Harm done accidentally or incidentally, whatever its scale, is not evil; it is misfortune, or negligence, or error — categories with their own moral weight, but a different weight.

Dangerous is capability without intent. A dangerous entity or situation carries a high probability of degrading agency, but no malice is required. Natural disasters, industrial accidents, the unintended fallout of benign projects — all dangerous, none evil. Danger is a fact about likelihoods; evil is a fact about aims.

The two dimensions are independent, which is where the taxonomy starts paying. An agent can be evil but not dangerous: someone who genuinely intends harm but lacks any practical means to inflict it is morally culpable and pragmatically harmless — malice without teeth. And capability without malice is everywhere: every driver on the highway is dangerous to you; almost none are evil.

A menace is both: deliberately malicious and sufficiently capable to pose a credible threat. This is the category that actually warrants fear, and it is rarer than either component alone. The practical payoff of keeping the axes separate is that they call for different responses. Culpability tracks intent — blame the evil, not the dangerous. Risk management tracks capability — guard against the dangerous, whatever their intentions. Confusing the two produces both of our characteristic errors: demonizing the merely dangerous, and indulging the not-yet-capable evil until it acquires means.

Comfort Is Not Care

Now put the definition to work on a case where it cuts against the prevailing culture rather than with it.

Contemporary moral sensibility treats discomfort as harm in itself and comfort as care in itself. That inversion follows directly from the confusion this chapter is built to dissolve — the confusion of emotional reaction with functional impact. Pain is not necessarily harm, and comfort is not necessarily care. The surgeon’s incision causes pain and heals; the sycophant’s flattery gives comfort and corrupts. Once harm is defined by capacity rather than feeling, the two cases sort themselves correctly, and a widely practiced form of “kindness” is exposed as its opposite.

Consider delusion. A delusion is not a harmless eccentricity; it is a systematic distortion of reality, a conceptual map that no longer corresponds to the terrain. An agent navigating by a false map has degraded capacity in exactly this chapter’s sense: its ability to choose and act effectively toward its own goals is impaired at the source, in the model of the world the choosing runs on. To indulge the delusion — to affirm, soothe, and politely acquiesce in it — is to reinforce the distortion, to strengthen the mental constraints that impede the agent’s action. One does not liberate a prisoner by decorating the cell walls in pleasant colors. Liberation requires breaking the lock.

Indulgence is therefore not neutral. It is complicity in the degradation of another agent’s capacity — and it cannot hide behind consent, because the agreement it obtains is manufactured by the very delusion it feeds. Consent must be informed; a “yes” produced by the false map is not the agent’s judgment but the distortion’s. The friend who supplies the addict with another dose betrays them. The partner who humors paranoid ideation betrays them. A society that cultivates fragile illusions into entrenched pathologies betrays its citizens. Each act feels gentle, and each communicates the same message: I value your immediate comfort above your long-term freedom; I prefer your passivity to your flourishing.

Genuine compassion is not sentimentality; it is resolve. It respects the other as an autonomous agent capable of engaging with truth. Indulgence condescends — you are too weak to endure reality — while confrontation, gentle or severe, affirms the opposite: you are strong enough to face it. And genuine compassion accepts a cost that false kindness never will. It risks rejection, conflict, resentment — the pains of the incision — for the sake of the other’s restored capacity. Only spurious compassion takes refuge in comfort. The test is the same as everywhere else in this chapter: not did it feel bad? but did it degrade or restore the agent’s power to pursue what it values? By that test, the flatterer harms and the honest friend heals, whatever the immediate feelings report.

One Definition, One Upgrade

The definition presented here — non-consensual degradation of an agent’s capacity to pursue its valued goals — is the canonical early form of this framework’s harm concept, and I should say plainly that it is not the final form. Part VI radicalizes it. When the ethics of viability arrives, harm gets reframed as the reduction of an agent’s viable futures, and imposed risk turns out to be harm in its probabilistic form — harm at the moment of decision, whether or not the bullet lands. That move, made in Risk Is Harm, extends this chapter rather than replacing it: the capacity definition survives as the special case where the degraded futures have already collapsed into the present, and every exclusion argued here — offense, rejection, envy — survives with it, because none of them reduces anyone’s viable futures either. One definition with an honest upgrade path, not three competing ones.

What is settled already, and will not move: harm is about agents, measured against their own valued goals, in terms of functional capacity, and filtered by consent. It is a fact about impact, not a verdict about wrongness — the verdicts come later, and they come out cleaner for having a clean fact to rest on.