Interiority Policing: When Governance Reaches the Source Rather Than the Act

When Governance Reaches the Source Rather Than the Act

Devin Arkema

Abstract. Some governance moves are not merely about how widely a rule reaches but about where in a person it lands. This paper develops a general theory of interiority policing: the structural shift in which regulatory attention extends from acts (what people do) to sources (what they imagine, intend, or dispositionally form). I define this shift as a category violation: a regime treats a source-condition of agency as though it were a conduct-event while preserving the consequences appropriate to conduct regulation. Drawing on a tightly bounded reading of the diagnostic field announced in Genesis 6:5 — “every intention of the thoughts of his heart” — and on three contemporary cases (predictive policing built on inferred mental states, radicalization-prevention programs that monitor disposition, and workplace and consumer-platform sentiment governance), I argue that interiority policing is not a quantitative escalation of conduct regulation but a structurally distinct governance type. Its signature has two parts. First, it expands the assessed field beyond the visible: the relevant data is no longer behavior but the formative substrate that produces behavior. Second, it makes defense structurally asymmetric: an agent cannot demonstrate non-intention or non-imagination in the way an agent can demonstrate non-performance of an act, and the standard moves available to a defendant under conduct law (denial of act, alibi, justification, mitigation by character) lose traction because the regulated phenomenon is the producing layer rather than the produced surface. I argue that this asymmetry of demonstration, rather than any particular abuse, is what makes interiority policing a category violation rather than an ethical excess. The paper closes by proposing a principled boundary: the capacity for refusal — the structural veto that converts a forming desire into a non-act — is the one interior signal a legitimate governance system must treat as inviolable. I operationalize this boundary through three minimal institutional tests: a conduct-rooted trigger requirement, a no-confession constraint, and a demonstrated-non-act requirement. Where regulation begins to subordinate refusal-capacity itself, governance has crossed from policing acts to policing the agent.

1. Introduction: Where the Rule Lands

Most arguments about how far a rule should reach are arguments about scale. Should the rule cover this class of persons or that one; should it apply to private as well as public actors; should it bind future generations as well as present ones. Those are real questions, but they are not the only questions of governance reach, and I am not, in this paper, going to argue about any of them. The question I want to develop concerns a different axis. Some governance moves are not really about how widely a rule applies. They are about where in a person it lands.

Take a perfectly ordinary criminal statute. The state forbids striking another person. The rule is a rule about an act. It tells me what I may not do. There is a substantial body of doctrine — the doctrine of mens rea, the doctrine of attempt, the doctrine of justification — that determines exactly which acts the rule covers and exactly which excuses block conviction. These doctrines have been contested for centuries, but they share a common shape. They take, as their object of regulation, an external occurrence — a striking, a taking, a killing, a saying — and they specify the mental conditions under which the occurrence will count as the kind of act the rule prohibits. The rule lands on conduct, even though it asks questions about intent on its way there.

Now consider a different kind of rule. Imagine that the state decided to forbid not striking but disposition to strike — that is, the formation in the imagination of a striking sufficient to predict it, before any external occurrence. Imagine, further, that the relevant evidence for whether one had formed such a disposition consisted not of conduct but of inferred mental states: scores from a behavioral risk model, transcripts of digital speech recompiled into a sentiment profile, declarations elicited in a screening interview. The rule still concerns striking. But the rule no longer lands on the act of striking. It lands on what produces the act — on the imaginative, dispositional, intentional substrate from which striking issues. I want a name for that move. I will call it interiority policing.

The phrase is meant in the broadest possible sense. By interiority I mean what an older theological and philosophical vocabulary called the heart, what an analytic philosophical vocabulary calls intention or disposition or second-order desire, and what contemporary surveillance practice calls behavioral signal or risk indicator. By policing I mean the regulatory function generally, not the institution of municipal law enforcement specifically. Possible instances of interiority policing, in the sense developed here, include: thought-crime statutes; predictive policing systems that allocate suspicion on the basis of inferred propensity; radicalization-monitoring programs that flag dispositional drift before any act has occurred; social-credit and reputational systems that score attitude rather than conduct; therapeutic regimes that reach the producing layer rather than the produced one; and the soft platform governance that classifies and intervenes on inferred sentiment. The list is meant as a taxonomy rather than a comprehensive case set. None of these regimes is identical to any other, and several have entirely different histories and justifications. What they share is a structural feature, and the paper develops that feature by working through three of these instances in detail — predictive policing, radicalization monitoring, and platform-mediated sentiment governance — chosen because they span the spectrum of formal coercion from state criminal authority through hybrid civilian-state programs to low-formal-coercion platform classification. (I avoid the unqualified phrase “non-coercive” for platform governance because workplace and consumer-platform regimes can be coercive in economic and infrastructural ways even when they carry no state coercion as such.) The other instances on the taxonomy are left for separate treatment, but the structural argument is intended to apply to all of them.

I want to argue that this shared structural feature is not merely a more invasive version of conduct regulation. It is a different kind of governance, with a different logic, and with a different set of failure modes, and I want to defend the claim that this difference is principled rather than merely scalar. The argument has three movements. First, in §2, I take a tightly bounded text — Genesis 6:5 — as a case study in how a governance move that announces interiority as its diagnostic field changes the regulatory game. I do not, here, treat the verse as authoritative; I treat it as an unusually clean instance of the move I want to analyze. Second, in §§3–5, I show that the same structural shift recurs in three contemporary regimes — predictive policing, radicalization monitoring, and platform-mediated sentiment governance — and I develop the formal asymmetry that distinguishes interiority policing from conduct regulation. Third, in §§6–8, I argue that the principled boundary at which interiority policing becomes a category violation, rather than merely an ethical excess, can be located by attending to one specific interior capacity — the capacity for refusal — that any legitimate governance system must treat as structurally inviolable. The conclusion is not that the interior is a permanent legal sanctuary; the conclusion is that one specific feature of interior life functions as a structural signal of agency, and that governance which subordinates that signal has crossed from regulating acts to regulating the agent.

I want to fix the central accusation in a single sentence before turning to the case. A category violation, in the sense I will use the phrase, occurs when a governance regime treats a source-condition of agency as though it were a conduct-event, while preserving the consequences appropriate to conduct regulation. The phrase has two halves. The first half — treating a source-condition as a conduct-event — names what is wrong with the diagnostic-field shift: the producing layer is not the kind of thing one can perform, refrain from, or stand accused of in the way that a conduct-event can be. The second half — preserving the consequences appropriate to conduct regulation — names why the shift matters institutionally rather than merely descriptively: the regime imports the apparatus of conduct law (detention, deprivation of access, classification with downstream effects) into a domain whose evidentiary form does not match the apparatus. This is the formal accusation I will defend, and the rest of the paper is an unpacking of its two halves and an argument for the boundary at which the violation becomes principled rather than rhetorical.

Three preliminary clarifications, the first of which I want to mark with some emphasis because it is the objection that legal-philosophy readers will press first. I am not arguing that mens rea — the mental component of criminal liability — is itself a form of interiority policing. Mens rea operates as a condition on whether an external act counts as a particular legal kind. It is the gateway through which the rule lands on conduct; it is not the rule itself. The point at which mens rea would become a form of interiority policing is the point at which mental states themselves become the prohibited thing, independently of any conduct they might condition. The criminal law has, on the classical doctrine, resisted that move; the diagnostic-field shift I will describe is the point at which the resistance has begun to fail. I return to this in §§5 and 7, but it is worth flagging at the outset that the argument does not depend on a dispute about mens rea. It depends on the more specific claim that a regulatory layer has been built up alongside the criminal law in which the act-conduct requirement no longer holds. Second, I am not arguing that attention to interior life is per se objectionable. A great deal of education, therapy, religious formation, and ordinary friendship consists in attention to interior life, and most of it is not policing in any sense. The structural feature I am isolating is policing — that is, regulatory action backed by enforcement asymmetry — directed at interior states. Third, the case I am building is not specifically liberal in its premises. It does not depend on Mill’s harm principle; it does not depend on Rawlsian neutrality; it does not depend on a strong claim about cognitive privacy as a value. It depends, instead, on an argument about what governance loses when its diagnostic field expands beyond the conditions under which defense against the rule remains possible. That argument can be entered from a number of directions. I will draw, particularly, on Hart (1968) on the relation of conduct to liability, on Foucault (1977, 1991) on disciplinary and pastoral power, on Arendt (1951, 1963) on the totalitarian penetration of the interior, and on the recent surveillance literature (Ferguson 2017; Zuboff 2019; Harcourt 2007) for contemporary instances. I will also draw, where relevant, on my prior work on agency, oversight, and refusal (Arkema 2025a, 2025b, 2025c, 2025d), since the argument I am making here grows directly out of that prior work and would be unintelligible without it.

2. The Diagnostic Field: A Reading of Genesis 6:5

I begin with one tightly bounded text, not because the argument depends on its authority but because the verse states with unusual compression the move I want to analyze. Genesis 6:5, in the standard English translations, reads: “The LORD saw that the wickedness of humankind was great in the earth, and that every intention of the thoughts of his heart was only evil continually” (Gen 6:5, NRSV). The narrative function of the verse is to motivate what follows — the divine decision to send the flood. The structural function, which is what I want to attend to, lies in the way the verse describes the diagnostic field on which that decision is taken. The clause “every intention of the thoughts of his heart” does not say that the wickedness of humankind was great because of the great number of wicked acts, although it does not deny that. It says that what was wicked was the formative substrate from which acts issue: “intention” (Hebrew yetser, the formative impulse), “thoughts” (machshevot), “heart” (lev — in biblical anthropology, the seat of will and reasoning, not merely affect). The diagnostic field on which the verdict is taken extends beyond outward conduct to the producing layer.

The structural reading I am taking from the verse is recognized in the major commentary tradition. The classic critical commentaries on Genesis converge on the observation that what is novel about 6:5 is not the severity of the verdict but the location of its diagnostic field: the assessment is described as reaching into the formative organ of conduct rather than alighting on the surface of conduct itself. Gerhard von Rad’s commentary on Genesis (von Rad 1972, ad loc.) treats the verse as a deliberate universalization of the diagnosis; Claus Westermann’s philological treatment (Westermann 1984, ad loc.) reads the construction — “every,” “only,” “continually” — as deliberately closing off any conduct-free remainder within which an alternative case could be assembled; Walter Brueggemann’s theological commentary (Brueggemann 1982, ad loc.) emphasizes that the chapter’s narrative response is keyed to the producing layer rather than to acts; Gordon Wenham’s philological treatment (Wenham 1987, ad loc.) develops the lexical and syntactic case for the same point. I draw on this material not to defend a theological reading but because the agreement among the major commentators allows me to take the structural point as established without arguing for it from scratch. The Hebrew vocabulary the verse deploys — yetser (formative impulse), machshevot (thoughts), lev (in biblical anthropology, the seat of will and reasoning) — names the formative process from which conduct emerges. The diagnostic field on which the verdict is taken extends beyond outward conduct to the producing layer.

Read as a piece of governance theory rather than as a piece of theology, the verse is an unusually clean exhibit of a move with a definite signature. The signature has three parts. First, the assessed field expands beyond what is visible. “Intentions” and “thoughts” and “the heart” are not features of conduct. They are features of the formative process from which conduct emerges. Second, the assessment is universal in scope: not “these intentions” but “every intention,” not “sometimes” but “continually.” The text deliberately closes off the standard defensive move available under conduct law: there is no remainder, no conduct-free interior space within which an alternative case could be assembled. Third, the response keyed to the diagnosis is not regulatory in the conduct sense but reconstructive in the formative sense — the response is to the producing layer, not to the produced acts.

The point I want to take from this is structural and limited. The verse does not show that interiority policing is wicked, nor does it show that interiority policing is justified when the assessor is omniscient. It shows that, when the diagnostic field is shifted from acts to sources, certain governance moves become possible that were not possible before, and certain defensive moves become impossible that had been possible before. The shift changes the regulatory game. The text presents that shift as belonging to a sovereign with cognitive access that no human institution has. Once that access is abstracted away, what remains is the formal logic of the shift itself — a logic that human institutions can attempt to instantiate, and increasingly do.

It is worth saying once more, in plainer terms, what role this section is playing in the paper. Genesis 6:5 is not the source of the norm I will defend. It is the first diagnostic model — the earliest exhibit I know of in which the assessed field is explicitly interior, the assessment is universal in scope, and the response is keyed to the producing layer rather than to acts. I do not need the verse to be true, I do not need it to be authoritative, and I do not need the reader to share its theological frame. I need only the structural reading: that the move from acts to sources, once made, changes what defense and assessment look like. The contemporary regimes I turn to next make the same move, under wholly different auspices, and that is the burden the rest of the paper will carry.

3. Three Contemporary Instances

If the formal logic just described were confined to one biblical episode, it would be of antiquarian interest only. It is not. The same shift recurs across at least three contemporary regimes, each with its own justification, history, and failure modes, and each marked by the same structural signature: the assessed field expands beyond visible conduct to the substrate that produces conduct.

Before turning to the three cases, one orienting note. The most direct instantiation of the category violation I have just defined is the pure thought-crime statute — the rule whose object is a mental state independent of any conduct it might condition. The pure thought-crime case has been resisted in legal theory precisely because it abolishes the act-conduct requirement entirely; that resistance is part of why the criminal law has not, at least in liberal-democratic jurisdictions, actually arrived at such statutes in their pure form. The three cases I develop below are not the most extreme instances of the diagnostic-field shift. They are the harder cases — the regimes in which the shift is happening inside institutions that still retain some conduct-attached doctrine, and in which the shift therefore has to be located analytically rather than read off the statute book. The pure thought-crime case stands as the clarifying pole; the three cases below sit on the spectrum between conduct regulation and that pole, at varying distances from it.

The first instance is predictive policing. The phrase covers a range of practices, including place-based forecasting (which neighborhoods are likely to see crime tonight), person-based forecasting (which individuals are likely to commit, or be victims of, crime in the coming period), and risk-classification systems that are used to inform pretrial detention, probation, and parole. The technical apparatus differs across these systems, but the regulatory move is the same: the system allocates state attention not to acts that have occurred but to dispositional inferences about acts that have not occurred. Andrew Guthrie Ferguson’s analysis (2017) makes the structural point clearly. Person-based predictive systems do not redescribe an existing pattern of conduct; they produce a new evidentiary kind — a risk classification, derived from data inputs that include but are not limited to prior conduct — and they license consequential state action on the basis of that classification. Bernard Harcourt (2007) had argued, a decade earlier, that the introduction of actuarial prediction into criminal justice represented a fundamental change in what criminal justice was for, since the categories of preventive risk are not the categories of moral responsibility. What both authors register, in slightly different terms, is the diagnostic-field shift. In its classical criminal-law register, the state presents itself as responding to conduct that has occurred. Even there, the picture is more layered than the self-presentation suggests: preventive detention, parole conditions, conspiracy doctrine, attempt liability, civil commitment, and prior restraint all complicate the picture. What is new is not that the state has begun to anticipate future conduct — it has long done so — but that an additional regulatory layer has been built up in which the substrate from which conduct might issue is itself the assessable thing, and in which the consequential institutional decision is conditioned on a dispositional inference rather than on a conduct event.

The second instance is radicalization monitoring. Counter-extremism programs in the United Kingdom (Prevent, and its statutory successor under the Counter-Terrorism and Security Act 2015), the United States (Countering Violent Extremism initiatives, in their various incarnations), and elsewhere in Europe and Australasia operate on a logic that is explicit about the diagnostic shift: the program is designed to intervene before any criminal act has been planned, on the basis of “indicators” that purport to identify dispositional movement toward extremist violence. Arun Kundnani’s extended critique (2014) of the radicalization-monitoring paradigm makes two structural points that bear on the present argument. First, the indicators are explicitly dispositional rather than conduct-based: changes in dress, in religious practice, in associative pattern, in expressed attitude. Second, the burden of proof structure available to a person flagged by the indicators is fundamentally different from the burden structure available to a person charged with a conduct offense, because the relevant evidentiary kind is not an event but a trajectory. To rebut a charge of conduct, one shows the conduct did not occur. To rebut a flag for dispositional drift, one would have to demonstrate the absence of a forming disposition — a demonstration with no clean evidentiary form.

The third instance is the softest of the three, and I want to present it as such. It is workplace and consumer-platform sentiment governance: monitoring systems that classify employee “engagement,” “attitude,” or “fit” on the basis of communication metadata and behavioral telemetry; and consumer-platform systems that classify users by inferred ideological or emotional profile in order to allocate access to features, content, distribution, or commercial opportunity. I am deliberately bracketing, here, the harder and more institutionally consolidated case of state-administered social-credit or reputational systems. Those exist (Creemers 2018; Kostka 2019), and they instantiate the same structural shift in a more coercive register, but the work this section needs to do is more cleanly done by the softer case, since the softer case shows that the diagnostic-field shift is not a property of state coercion. Shoshana Zuboff’s analysis of “behavioral futures” (2019) is the most-cited treatment of the underlying logic. Zuboff’s claim is that the platform economy has shifted from the extraction of revealed-preference data, on which it could only act after the fact, to the extraction and modification of dispositional data, on which it can act before the fact. The key term in Zuboff’s analysis, for present purposes, is “modification” — the platforms are not merely observing dispositions in order to predict conduct but altering dispositions in order to shape conduct. The diagnostic field of regulation is, again, the producing layer, not the produced surface. That this happens without the formal coercion of the state is part of the point. Workplace and consumer-platform regimes can still exert significant economic and infrastructural pressure on those they classify; the point is only that the structural feature is generic and does not require the particular institutional auspices of criminal law to appear.

I want to be clear about what I am claiming and what I am not claiming. I am not claiming that these three regimes are equivalent. They are not. Predictive policing carries the coercive power of the state. Radicalization monitoring is a hybrid: a state program with intermediating civilian referral institutions. Workplace and consumer-platform sentiment governance, the softest of the three, generally does not carry state coercion at all, even though it can carry significant economic and infrastructural force. Each of these regimes has its own moral character and its own history, and a complete analysis of any one of them would require attention to its specific institutional features. The claim I am making is more limited. The three regimes share a structural signature — the diagnostic field has been shifted from acts to sources — and that shared signature is not coincidental. It is the formal feature that I have been trying to identify. The fact that it can be instantiated under such different institutional auspices, with such different distributions of formal coercion, is itself part of the argument: the shift is not the property of any one institutional form. It is a generic move that any sufficiently capable governance system can make once the relevant inferential capacity is available.

4. The Asymmetry of Defense

Why is the diagnostic-field shift not merely a more invasive version of conduct regulation? The answer turns on a feature that becomes visible once one attends to the structure of defense rather than the structure of accusation. Under conduct law, certain defensive moves are intelligible because the regulated phenomenon is an event in the world. The defendant can deny that the event occurred (the act did not happen, or did not happen as alleged); can deny that the event was of the relevant legal kind (the conduct was not theft because the property was not another’s); can offer justification (the conduct was prohibited but warranted in the circumstances); can offer excuse (the conduct was prohibited but the agent was not in the relevant sense responsible for it); or can argue that the conduct, while prohibited and unjustified and inexcusable, ought to be treated leniently in light of the agent’s broader character or circumstance. These defensive moves are not exhaustive, but they share a common feature: they have an evidentiary form. There is something the defendant can do — a witness, a document, a measurement, a chain of inference — that bears intelligibly on whether the rule has been broken.

This evidentiary form depends on a structural feature of the regulated phenomenon: the regulated phenomenon is bounded. The act either occurred or did not. The taking either was of another’s property or was not. The killing either was in self-defense or was not. Hart’s analysis of the relation between conduct and liability (Hart 1968) takes precisely this boundedness as the condition under which a rule of conduct can be defensible at all: the rule is fair if and only if the agent can know, in advance, what conduct it forbids, and can know, after the fact, whether the conduct alleged has been performed. Simester and von Hirsch (2011), in their later treatment, make the point in a slightly different vocabulary by tying the legitimacy of criminalization to the agent’s capacity to recognize and avoid the act in question. In both cases, the structural condition for the rule’s legitimacy is that the regulated phenomenon admits of a clean evidentiary handle.

Interiority policing has no such structural condition. The regulated phenomenon — “intention,” “disposition,” “attitude,” “trajectory” — is unbounded in two distinct senses. First, it is unbounded in time: a disposition is not an event but a tendency, and tendencies do not begin and end at instants the way acts do. Second, it is unbounded in form: the relevant evidence is no longer an event in the world but an inference about the producing process, and there is no finite list of features that exhausts what counts as evidence of a forming disposition. The result is that the standard defensive moves lose their grip. The defendant cannot deny the event because there is no event to deny. The defendant cannot offer justification because there is no act to justify. The defendant cannot offer excuse because excuses are about acts, and the regulated phenomenon is not an act. What is left is something closer to a defense of one’s general formative life — an account, addressed to the regulator, of what one is in the process of becoming. That defense does not have a clean evidentiary form, because the producing layer does not. One could put it more compactly than the careful version: one cannot not intend, one cannot not imagine, and the standard defensive moves available under conduct law do not survive that fact.

I want to put one point about that line more carefully than the punch of the phrasing demands, since a hostile reader will press on it. The claim is not that an agent has no capacity to redirect imagination or to refrain from cultivating an intention; agents plainly do have such capacities, and a great deal of moral life consists in their exercise. The claim is the narrower one that the agent cannot demonstrate non-intention or non-imagination in the way an agent can demonstrate non-performance of an act. The non-act has an evidentiary handle — a place, a time, a witness, a record — that the non-intention does not. The asymmetry I am isolating is the asymmetry of demonstration, not the asymmetry of capacity. An agent can be the locus of disciplined interior formation while still being unable to bring that formation into a regulator’s evidentiary form on the regulator’s schedule, and that inability is what the rest of this section is about.

I want to call this the asymmetry of defense. It is the formal feature that distinguishes interiority policing from conduct regulation, and it is the feature that, on my view, gives content to the category-violation diagnosis I introduced in §1. Recall the definition: a regime commits a category violation when it treats a source-condition of agency as though it were a conduct-event, while preserving the consequences appropriate to conduct regulation. The asymmetry of defense is what makes the first half of that definition bite. Under conduct regulation, the agent and the regulator face the rule on roughly comparable epistemic terms: both can, in principle, ascertain whether the conduct in question has occurred. Under interiority policing, the agent and the regulator do not face the rule on comparable terms. The regulator has a model — a risk score, a sentiment classification, a trajectory inference — that purports to characterize the agent’s producing layer. The agent has nothing comparable. The agent cannot produce a counter-model of equal evidentiary standing because there is no shared evidentiary form within which the model can be contested. The agent’s denial — “I am not in fact moving toward extremism,” “I am not in fact developing the disposition to violence,” “I am not in fact disengaging from the workplace” — is structurally incommensurable with the regulator’s assessment, because the assessor has built the assessment from data of a kind the agent cannot match.

Three implications of the asymmetry are worth stating explicitly. First, the asymmetry is not contingent on the regulator’s having gotten something wrong; it is a property of the regulatory structure even when the regulator has gotten everything right. A correctly inferred disposition is still inferred from a kind of evidence to which the agent has no equivalent counterclaim. Second, the asymmetry is not removed by giving the agent procedural protections that are appropriate to conduct regulation. A right to know the charges, a right to confront the evidence, a right to counsel — all of these presuppose an evidentiary form within which the contest is intelligible. They do not generate that form when the regulated phenomenon does not admit of one. Third, the asymmetry compounds across cases. A risk classification produced by interiority policing in one regime can become an input into another regime’s classification, and the further downstream the classification travels, the more it functions as an evidentiary kind in its own right rather than as a derivative inference. The agent’s opportunity to contest the inference, already structurally limited at the first step, becomes vanishingly small at later steps.

One further clarification, since readers familiar with the predictive-policing and radicalization-monitoring literatures will expect it: false positives intensify the injustice of interiority policing, but they are not the source of the category problem. The argument I have just made about the asymmetry of defense holds even when the regulator’s inference is correct. A correctly inferred disposition is still inferred from a kind of evidence to which the agent has no equivalent counterclaim; the regime still imports the apparatus of conduct law into a domain whose evidentiary form does not match the apparatus. The empirical question of how often these systems get the inference right is therefore not the question on which the present argument turns. The structural objection survives even high-accuracy inference, because what makes interiority policing a category violation is not its error rate but the form of regulation it instantiates.

Take the strongest version of the regulator’s case: suppose the inference is perfectly accurate — the dispositional risk real, the trajectory correctly characterized, the model error-free. Why should the structural objection survive? Because it is not about the accuracy of the inference but about the form of the regulatory decision that follows from it. A perfectly accurate inference about a forming disposition still produces a classification of an evidentiary kind to which the agent has no commensurate counterclaim; the agent’s self-understanding is not in the relevant evidentiary form. High accuracy does not produce evidentiary symmetry; only structural redesign does. The procedural complaint and the structural complaint are different complaints, with different conditions of dissolution: under perfect inference the procedural complaint dissolves, but the structural complaint does not, because the agent is still being characterized in a form against which they have no commensurate way to bring evidence. If anything, perfect inference exacerbates the problem: a system capable of fully legibilizing the producing layer is precisely the system in which the asymmetry of demonstration is most severe.

5. Two Genealogies

It is sometimes objected that the move I am calling interiority policing is not new — that the criminal law has always cared about the mind, that confessional and pastoral practices have always pressed inward, that no era has ever entirely confined regulation to the bare external act. The objection is partly correct, and it is worth taking seriously, because the historical situation is more layered than a clean origin story would suggest. There are two genealogies one can trace. The first runs through the theological and confessional tradition; the second runs through the secular legal and governmental tradition. Both feed into the contemporary regimes I described in §3, and the way they feed in is distinct.

The first genealogy begins, for our purposes, with Augustine. The interiorization of moral assessment in the Latin Christian tradition is not original to Augustine — it has Hellenistic, Stoic, and biblical antecedents — but Augustine is the figure through whom the interiorization becomes architectural. The famous self-examination of the Confessions is structured around the proposition that the relevant self is not the public self of acts but the inward self of intention, of disordered love, of pre-volitional formation (Augustine 1991, X.viii–xxxv). The diagnostic field for moral assessment, in Augustine’s practice, is the heart — and the corresponding pastoral practice, formalized in the medieval institution of auricular confession, requires the penitent to disclose not merely acts but the formative substrate from which acts proceed. Charles Taylor (1989) reconstructs this development as the gradual emergence of “radical reflexivity” as a characteristic shape of Western interior life: the self becomes a reachable object of one’s own assessment, and through pastoral mediation, of others’.

I want to be careful about what this genealogy does and does not establish. It does not establish that the contemporary regimes I described in §3 are descendants of Augustinian pastoral practice in any substantive sense; that would be a far stronger historical claim than the evidence supports. It does establish that the practice of interior assessment, as a structural form, has a long and well-documented history in Western institutions, and that the contemporary regimes are not the first to make their diagnostic field the producing layer rather than the produced surface. What is new in the contemporary regimes is not the move but the apparatus: pastoral interior assessment was bounded by the consent of the penitent and by the confessional seal; contemporary interiority policing is conducted at scale, without confession, without consent in any meaningful sense, and with the inferred classifications routed into consequential institutional decisions. The form of the move is old. The conditions under which it is exercised are new.

The second genealogy runs through the secular legal and governmental tradition, and is more complicated, because the legal tradition has always required some form of mens rea — some attention to the agent’s mental state — as a condition of liability. There is a tempting reading on which mens rea simply is interiority policing, and on which the move I have been describing has been at the center of criminal law from the beginning. The reading is wrong, and it is important to see why. Hart’s treatment of mens rea (Hart 1968) makes the structural point: mens rea operates as a condition on whether an external act counts as a particular legal kind. It is not the act that is the rule’s object; it is the conjunction of act and mental state. In particular, mens rea cannot, on the classical doctrine, supply the act it conditions. There must be a conduct-event for the mental state to be a condition on. The doctrine of attempt extends the conduct requirement somewhat — there must be some act-toward, even if the consummated act has not occurred — but the requirement is still that the mental state attach to a conduct event of some kind. Where the criminal law has experimented with pure mental-state offenses, the experiments have been strongly resisted in legal theory, precisely on the ground that they violate the act-conduct requirement.

What has happened, in the move toward interiority policing as I have described it, is not that mens rea has expanded but that the act-conduct requirement has been weakened, and a new evidentiary kind — risk classification, dispositional inference, trajectory model — has begun to function as a substitute for the conduct event. Foucault’s analysis of disciplinary and pastoral power (Foucault 1977; 1991) is useful here precisely because it is not a critique of mens rea. Foucault does not argue that the criminal law has always been about the soul. He argues, instead, that an additional regulatory layer — disciplinary, pastoral, and governmental — has been built up alongside the classical criminal law, and that this additional layer takes the producing-process of conduct as its object, while the classical law continues to take conduct as its object. The contemporary regimes I described in §3 are best understood not as continuations of mens rea doctrine but as instantiations of the disciplinary-pastoral layer at machine scale. The diagnostic field shift is not buried in the long history of criminal liability. It is a feature of the additional layer that has been constructed alongside criminal liability and that increasingly competes with it for regulatory authority.

What the phrase “machine scale” is doing here deserves more specification than I have so far given it, since the normative weight of the genealogy turns on it. Four features of machine-scale implementation are, on my view, what shift the structural transformation from a difference of degree to a difference of kind. The first is permanence: classifications, once produced, persist beyond the context of their production and accumulate into a record that travels with the subject. The second is opacity: the inferential apparatus that produces the classification is, in the typical case, not inspectable by the subject and often not inspectable even by the institution deploying it. The third is compounding: a classification produced in one regime becomes an input into another regime’s classification, and the further downstream the inference travels, the more it functions as an evidentiary kind in its own right rather than as a derivative inference. The fourth is the loss of the consent-and-confidentiality structure that bounded earlier forms of pastoral interior assessment: the auricular confessor was bound by the seal of the confessional and by the limits of what the penitent had voluntarily disclosed; machine-scale interiority assessment has no such bounds. None of these features is unique to machine-scale implementation, but their combination is, and it is the combination that produces the categorial shift the genealogy is attempting to mark.

Hannah Arendt’s analysis of totalitarianism (Arendt 1951; 1963) provides the third element of this genealogy. Arendt is sometimes read as if her central claim were that totalitarian regimes are simply more brutal than other regimes. Her actual claim is more specific. What distinguishes totalitarianism, on her account, is that it does not stop at the regulation of public conduct: it penetrates the interior, demanding not merely outward compliance but inward conviction. The terror of totalitarianism is, in Arendt’s description, addressed at the producing layer of personality itself. I cite Arendt not because the contemporary regimes I have described are totalitarian in her sense — they are mostly not — but because Arendt identifies, with great precision, the formal point at which a regime crosses from regulating acts to regulating agents. The point is not the introduction of mens rea. The point is the point at which inward formation itself becomes the assessable thing. The regimes I described in §3 do not, in general, demand inward conviction. They do classify inward formation. Whether the classification is benign, in any given case, depends on what is done with it. The structural risk Arendt identified does not. It attaches to the classificatory move itself.

6. The Refusal Principle

If the diagnostic-field shift is a category-distinct governance move, and if its asymmetry of defense is the structural feature that makes it problematic, what is the principled boundary at which interiority policing becomes more than an ethical excess? I want to argue that the boundary can be located by attending to one specific interior capacity: the capacity for refusal. The argument has two steps. The first step is to identify what is structurally distinctive about refusal. The second step is to argue that this distinctive feature gives refusal a privileged status as a governance signal — and therefore that any regime that treats refusal-capacity itself as an object of policing has crossed from acts to agents in a more than metaphorical sense.

On the first step, I will draw on the line of work I have been developing in earlier papers (Arkema 2025a; 2025d), but the argument I am making here can be entered without those papers in hand. Refusal, as I am using the term, is not the same as the second-order endorsement that Frankfurt (1971) made central to his account of personhood. Frankfurt’s analysis identified the capacity to want what one wants — to take a stance toward one’s own first-order desires — as the structural mark of agency. Refusal is a related but distinct capacity. It is the capacity to interpose a structural veto between a forming desire and an act. It is the no that is uttered, sometimes wordlessly, in the moment between an emerging motion and its execution. Frankfurt’s endorsement looks ahead from a stance of evaluation; refusal looks back at a forming motion and arrests it. The two capacities are related — refusal can become endorsement, and endorsement can fail to refuse — but they have different structural signatures and different times of operation. I do not, here, take a position on the substantial literature that has developed Frankfurt’s account in directions other than the one I am pursuing — Bratman’s planning theory of intention (Bratman 1987), Velleman’s constitutive account of agency (Velleman 1989, 2000), and the considerable secondary literature responding to both. The argument of this paper does not depend on adjudicating those developments; it depends only on the narrower claim that refusal-capacity is a structurally distinct capacity from second-order endorsement, and that the structural distinctness has consequences for governance theory.

Why does refusal have a privileged status as a governance signal? Two reasons. First, refusal is the one interior capacity whose exercise is, by its own nature, observable to the agent and only to the agent. The exercise of refusal does not produce an external event. It produces the absence of an event that would otherwise have occurred. The non-act that issues from a successful refusal is, from the outside, indistinguishable from the absence of any forming desire. This is not a contingent epistemic fact. It is a property of the structural role refusal plays in the production of conduct. Second, refusal is structurally prior to the other interior capacities in the sense that, without it, those other capacities lose their character as capacities of an agent. The argument is not merely that an agent who has lost the capacity to refuse has lost something important; the argument is that what looked like the other capacities collapse into something else when refusal is gone. Endorsement that cannot be withheld is not endorsement; it is compliance. The structural difference is this: an endorsement is a stance one takes that one might have not taken. If the agent has lost the structural capacity to interpose a veto between forming desire and act, the apparent endorsement disappears as a distinct moment in the production of conduct; what remains is the desire and the act, with the “endorsement” riding along as epiphenomenal commentary. The same logic applies to imagination. Imagination that cannot be checked against refusal is not imagination but automatism, because imagination is the structured entertaining of possibilities, and entertaining is, formally, a stance from which the agent can decline to proceed. Without refusal, what looks like imagination is just the producing layer producing. The argument generalizes. Refusal is not the most important interior capacity among others of the same kind; it is the structural condition under which the others count as capacities of an agent at all.

Both points have been made in different vocabularies in earlier work. The first appears in Arkema (2025d), in the analysis of free won’t as the foundational signal of agency. The second appears in Arkema (2025c), in the discussion of the recursive collapse that occurs when a system’s revision capacity is eliminated while behavior continues. What is new in the present argument is the application of these points to the boundary between conduct regulation and interiority policing. If refusal has the structural status I have just described, then a governance regime can — at least in principle — be assessed by the kind of relation it sustains to refusal-capacity. A regime that addresses itself to acts allows refusal to function: the agent’s capacity to interpose a veto between forming desire and act is preserved, and the regulator’s task is to assess what occurred when refusal failed or was not deployed. A regime that addresses itself to dispositions, sources, intentions, and trajectories — that is, a regime of interiority policing — operates on the producing layer, prior to refusal. Such a regime does not, on its face, abolish refusal-capacity; it can coexist with it. But it can also subordinate it.

What does it mean for interiority policing to subordinate refusal-capacity? It means that the regime’s assessments are taken to be authoritative independently of whether the agent has refused the disposition the regime has classified. The classification is read off the producing layer. The agent’s capacity to interpose a veto — to refuse the desire as it forms, to make the disposition not issue in act — is, on the regime’s logic, evidentially irrelevant, because the regime is assessing the producing layer rather than its outputs. The classification stands whether or not refusal is exercised. Where this is the structure of the regime, refusal-capacity has been subordinated: the agent retains the capacity, but the capacity does not register in the regulatory game. It cannot be deployed as a defense, because the defendant under interiority policing is not being charged with conduct.

I want to call this the Refusal Principle. The Refusal Principle states: any legitimate governance regime that takes interior life as a regulatory object must structure its assessments so that refusal-capacity remains a defense. Where the assessment is read off the producing layer in a way that makes refusal evidentially irrelevant, the regime has crossed a category boundary — it is no longer policing what people do, or even what people will do, but policing what people are. The Principle does not say that interior life is a permanent legal sanctuary. It says that one specific feature of interior life — the capacity to interpose a veto between forming desire and act — has the structural status of a signal that any legitimate regime must read. To read it requires that the assessment have a place for it. Interiority policing, as I have defined it, has no such place. That is the formal feature on which the principled objection rests.

6.1. Minimal Institutional Tests

The Refusal Principle, as I have stated it so far, is a structural minimum: refusal-capacity must remain available as a contestable signal in the regime’s assessment. A reasonable objection — and one I want to take seriously rather than leaving for the §7 catalogue — is that this is too abstract to do governance work, and that any regime can claim to satisfy it post hoc. To answer the objection, the Principle has to be operationalized: it has to specify the kind of regulatory architecture that actually preserves refusal as a defense, rather than the kind that gestures at refusal while building assessments that ignore it. I want to propose three minimal institutional tests. They are not jointly sufficient for the legitimacy of a regime that takes interior life as a regulatory object — there are doubtless others — but each is, on my view, individually necessary. Where any of the three fails, the regime has subordinated refusal-capacity in the sense developed above.

First, the conduct-rooted trigger requirement. A regime that takes dispositional or trajectory inference as input may not act on the inference alone; the consequential institutional decision must be tied to a conduct event of some kind, even a minimal one. The trigger must be conduct-rooted in the sense that it is contestable as an event or omission attributable to the subject, not merely a model-derived classification assembled from diffuse telemetry. This rules out the obvious evasion in datafied environments: a system cannot satisfy the requirement by declaring that all data ultimately originates in conduct, since what does the regulatory work in such systems is the model-derived classification rather than any specific event the subject could identify, perform, refrain from, or contest. The point of the requirement is not that the conduct event is the whole of what the regime regulates — the regime may, by design, also be tracking and modeling the producing layer — but that no consequence may attach without a conduct-rooted trigger that brings the assessment back into commensurable evidentiary form. A further constraint follows from the same logic: the trigger must be materially connected to the asserted risk, not merely any act or omission attributable to the subject. A regime cannot satisfy the requirement by pointing to a trivial conduct hook unrelated to the dispositional inference it has drawn — clicking a link once, attending a single lecture, sending one message — and then importing the consequences appropriate to a serious dispositional finding. The act-conduct trigger and the asserted risk must hang together; absent that connection, the “trigger” is doing only ceremonial work and the regime has not, in substance, satisfied the requirement. This is recognizable as a weak version of the act-conduct requirement that the criminal law has long maintained, transposed into the regulatory context. Predictive policing systems that condition police contact on a generic risk classification fail the test. Workplace sentiment systems that condition adverse employment action on a “disengagement” score fail the test. Systems that route a risk classification into investigative attention without immediate consequential action — and that act only when an investigation, conditioned on a conduct trigger, produces independent conduct-rooted evidence — pass it.

Second, the no-confession constraint. A regime that takes interior life as a regulatory object may not, in the course of contesting a classification, force the subject to narrate or confess interior life as the price of the contest. The subject must be able to challenge the classification without being placed in the position of having to perform, on the regulator’s schedule, a defense of what they are becoming. The constraint is the structural cousin of the legal-procedural protection against self-incrimination, but it is broader: it covers any institutional procedure that conditions remedy on the production of an interior account the subject would not otherwise be required to provide. Radicalization-monitoring programs that ask the flagged subject to demonstrate their non-extremist disposition through expressed avowal fail this test. Programs that test the underlying conduct-rooted indicators directly, and that allow the subject’s contestation to proceed in conduct-rooted form, pass it.

Third, the demonstrated-non-act requirement. A regime that takes dispositional inference as input must allow that the demonstrated absence of action — the documented withdrawal, the de-escalation, the visible refusal — counts against the classification rather than being treated as evidentially irrelevant or as evidence of further concealment. This is the test that most directly tracks the Refusal Principle, because it is the test that requires the regime to read refusal when refusal is exercised. A regime that treats every absence as either compliance or concealment, with no evidentiary weight assigned to the demonstrated non-act, has built a procedure within which refusal cannot register. The structural consequence is that the agent retains the capacity to refuse but loses the standing to be heard as having done so. The demonstrated-non-act requirement protects against this consequence by requiring that refusal-evidence be admissible in the regime’s own evidentiary form.

Each of these three tests is austere. None of them, by itself, decides whether a regime is otherwise just. What they do is convert the Refusal Principle from a moral axiom into a usable diagnostic. A regime that satisfies all three has preserved the structural minimum that makes the agent’s presence in the regulatory game intelligible. A regime that fails any one of them has, to that extent, crossed the boundary I have been trying to specify — it has begun to police the agent rather than what the agent does.

7. Objections and Replies

Several objections to the argument as I have stated it deserve direct attention. I will take five. The first is that I have understated the legitimacy of mens rea. The second is that I have overstated the asymmetry of defense. The third is that radicalization monitoring is, in fact, justified by the harms it prevents, even if it has the structural feature I have described. The fourth is that the institutional tests I propose in §6.1 are already widely satisfied — that practitioners in existing programs already weight non-acts as exculpatory and the test is therefore weaker than the paper presents. The fifth is that the Refusal Principle is too thin to do real work — that it merely renames the problem rather than solving it.

On mens rea. The objection is that the long-standing requirement that criminal liability include a mental component shows that the law has always policed the interior, and that the diagnostic-field shift I have described is the continuation of a tradition rather than a category-distinct move. I have already argued, in §5, that this is the wrong way to read the doctrinal history. Mens rea operates as a condition on whether an external act counts as a particular legal kind; it is not, on the classical doctrine, an independent target of regulation. The pure mental-state offense — the offense in which mental state is the prohibited thing, independent of any conduct it might condition — has been resisted in legal theory precisely because it abolishes the act-conduct requirement that mens rea presupposes. The diagnostic-field shift I have described is not the expansion of mens rea but the weakening of the act-conduct requirement. The two are different moves with different consequences. Where contemporary regimes have produced something that looks like the convergence of the two moves — pretrial risk classification, for example — the convergence is, on my view, the relevant problem to address.

On the asymmetry of defense. The objection is that I have overstated the absence of evidentiary form for inferences about disposition. Defendants in many legal contexts have made arguments about character, propensity, and disposition; the law has long had to address evidentiary questions about mental life; and the asymmetry I have described is therefore not as sharp as I have suggested. I accept that the asymmetry is not absolute. There are evidentiary practices for assessing disposition — character testimony, expert psychiatric evaluation, prior conduct as predictive of present conduct — and these practices have a long history. The relevant point is not that the asymmetry is total but that it is structural. The evidentiary practices that exist for assessing disposition all have, as their ultimate evidentiary object, conduct of some kind: prior acts, expressed statements, observable behavior, the patterns from which dispositions can be inferred. They are conduct-rooted inferences about disposition. What is new in the contemporary regimes I described is not that they make inferences about disposition but that the inferential apparatus has been disconnected from the conduct-rooted base. The disconnection is what produces the asymmetry. Where the inference is rooted in observable conduct, the agent has, at least, the conduct as a contestable evidentiary object. Where the inference is rooted in data the agent cannot produce a counter-version of — proprietary risk-model outputs, behavioral metadata aggregations, sentiment classifications derived from inputs the agent cannot inspect — the contestability is lost. The point of the asymmetry argument is that this loss is structural, not contingent.

On radicalization monitoring. The objection is that, even if radicalization-monitoring programs have the structural feature I have described, they are nonetheless justified by the harms they prevent — and that any complete account of governance reach must weigh those harms in the balance. I accept the importance of the question. The point I want to make is that the structural feature I have described is logically prior to the harm-prevention question. The fact that a regime would prevent serious harm does not, by itself, dispose of an objection that the regime crosses a category boundary in the structure of governance it employs. Compare an analogous case: the fact that a regime of confession would prevent serious harm does not dispose of an objection that the regime violates the structural conditions under which assessments are intelligible to the assessed. One must answer the structural objection on its own terms. My argument is that radicalization monitoring, as currently practiced, does not adequately answer the structural objection — that its evidentiary apparatus produces inferences that the monitored cannot contest in the form of the inference itself, and that the consequential institutional decisions taken on the basis of those inferences are taken in a regulatory frame that has not been redesigned to accommodate the asymmetry of defense. The harm-prevention question, however serious, cannot dispose of the structural question. It can, at most, supply a reason to think the structural question is worth solving rather than abandoned.

On the practitioner reply that the institutional tests are already widely satisfied. A reader familiar with the actual operation of programs like Prevent in the United Kingdom, or with the various Countering Violent Extremism initiatives in the United States and elsewhere, will note that several of these programs do explicitly treat behavioral de-escalation as exculpatory and weight it in classification outcomes. The objection that follows is that the demonstrated-non-act requirement is more widely satisfied than the §3 presentation suggests, and that practitioners would say of the third test, “we already do this.” The reply has a definite shape. The third test does not require merely that the regime sometimes weight non-acts. It requires that the demonstrated non-act be admissible in the regime’s own evidentiary form as a counter to the original classification. In practice, what most de-escalation-weighting regimes do is treat de-escalation as a partial mitigating factor that adjusts the consequences attached to a sustained classification, rather than as evidence that the original classification was wrong. The classification persists; the consequences soften. That is not the same as making refusal evidentially operative against the inference, and it does not satisfy the third test. The diagnostic question to put to a practitioner is this: when a flagged subject demonstrates de-escalation across the relevant indicators, does the system reclassify them as not-flagged, or does the flag persist with reduced consequences? In most existing programs, it is the latter. The flag, once entered, is a difficult thing to revoke even on extensive contrary evidence, because the regulatory architecture treats the upstream inference as the evidentiary fact and the downstream conduct as modulating but not overriding it. Where this is the architecture, the regime has not made refusal evidentially operative; it has made refusal cost-mitigating. The third test fails. The reply, then, is not that practitioners are mistaken to weight non-acts — the weighting is good as far as it goes — but that the practitioner who points to the weighting as evidence of compliance with the third test has confused mitigation of consequences with reclassification of inference. The two are structurally distinct, and the test requires the second.

On the Refusal Principle. The objection is that the Principle is too thin to do work — that the requirement to leave room for refusal-capacity in the regime’s assessment is so abstract that any regime can claim, post hoc, to satisfy it. I think this objection has bite, and I want to acknowledge what the Principle does and does not deliver. The Principle does not, by itself, generate a complete theory of legitimate governance reach. It does not specify which inferences about disposition are permissible and which are not, and it does not draw a bright line between regulation that respects the producing layer and regulation that subordinates it. What the Principle delivers is a structural test that any candidate regime must pass: there must be a place, in the regime’s assessment, for refusal as a contestable signal. Concretely: the regime must allow that the agent’s exercise or non-exercise of refusal-capacity bears on the regulatory determination, and it must build evidentiary architecture under which refusal can be brought into evidence. A regime that claims to satisfy the Principle but provides no such architecture has not, in fact, satisfied it; the claim is hollow, and the hollowness is detectable by examining the regulatory procedure rather than its rhetoric. The Principle is, in this respect, like other thin-but-binding governance constraints in the tradition — like Rawls’s publicity condition (Rawls 1993), or like the more recent corrigibility conditions developed in the AI-governance literature (Arkema 2025b). Such constraints do not specify everything one wants to know about a legitimate regime. They specify the structural minimum below which legitimacy fails.

8. Conclusion: Refusal as the Boundary of Legitimate Governance

The argument I have developed in this paper has had a deliberately narrow target. I have not argued that interior life is a permanent legal sanctuary, or that no regulation may attend to dispositions, intentions, or formative process. I have argued, instead, that the move from regulating acts to regulating sources is a structurally distinct governance type, that its signature is the asymmetry of defense produced when the assessed field expands beyond the conditions under which the rule can be intelligibly contested, and that the principled boundary at which the move becomes a category violation can be located by attending to refusal-capacity as the one interior signal a legitimate regime must allow to remain operative.

Three implications follow. First, the diagnostic-field shift is not a feature unique to one historical moment, one ideology, or one institutional form. It is a generic move that any sufficiently capable governance system can make, and the conditions for making it have proliferated. Predictive policing, radicalization monitoring, and platform-mediated sentiment governance are not the last regimes that will instantiate the structural feature I have described. They are early instances. The framework I have offered is intended to give analytic purchase on instances that have not yet been built. Second, the asymmetry of defense is not removable by procedural improvements that leave the diagnostic field intact. A regime can be made more transparent, more contestable in form, more rights-respecting in surface, while remaining a regime that addresses itself to the producing layer rather than to acts; and to the extent that it does, the structural objection survives the procedural improvement. The improvement is not by itself sufficient. What is required is a structural redesign that restores the agent’s capacity to face the rule on commensurable epistemic terms. Third, and finally, the Refusal Principle gives one specific shape to what such a structural redesign must preserve. It does not specify the shape uniquely. It specifies the minimum: refusal must remain available as a contestable signal in the regime’s assessment. Where it is not, the regime has crossed from policing what people do to policing the agent who would otherwise be the locus of doing.

A brief word about how this paper fits the broader research program. The argument here draws together three earlier moves I have made elsewhere: that refusal has a structurally privileged role in the constitution of agency (Arkema 2025d); that the legitimacy of governance over highly capable systems depends on the maintenance of corrigibility (Arkema 2025b); and that the agents who are subjects of governance are not bounded prior to their entanglements with the systems that govern them (Arkema 2025c). Interiority policing is the case in which the producing layer of agency has been treated as available for assessment by external regimes that have no commensurate capacity to be revised by it, and in which refusal — the structurally privileged signal — has been quietly subordinated to inferences from data the agent cannot contest. The diagnostic field has shifted, the regulatory game has moved with it, and the philosophical work that remains is to specify, in detail, what the agent’s structural standing must be when the rule lands not on what the agent has done but on what the agent is in the process of becoming.

References

Arendt, Hannah. 1951. The Origins of Totalitarianism. New York: Harcourt, Brace.

Arendt, Hannah. 1963. Eichmann in Jerusalem: A Report on the Banality of Evil. New York: Viking.

Arkema, Devin. 2025a. “Conviction Under Revisability.” PhilArchive. https://philarchive.org/rec/ARKCUR.

Arkema, Devin. 2025b. “From Static Oversight to Distributed Corrigibility.” PhilArchive. https://philarchive.org/rec/ARKFSO.

Arkema, Devin. 2025c. “The Entanglement You Were Already Inside.” PhilArchive. https://philarchive.org/rec/ARKTEY.

Arkema, Devin. 2025d. “Free Won’t vs Free Will.” PhilArchive. https://philarchive.org/rec/ARKFWV.

Augustine. 1991. Confessions. Translated by Henry Chadwick. Oxford: Oxford University Press.

Bratman, Michael E. 1987. Intention, Plans, and Practical Reason. Cambridge, MA: Harvard University Press.

Brueggemann, Walter. 1982. Genesis. Interpretation: A Bible Commentary for Teaching and Preaching. Atlanta: John Knox Press.

Creemers, Rogier. 2018. “China’s Social Credit System: An Evolving Practice of Control.” Journal of Contemporary China 27 (114): 824–841.

Ferguson, Andrew Guthrie. 2017. The Rise of Big Data Policing: Surveillance, Race, and the Future of Law Enforcement. New York: New York University Press.

Foucault, Michel. 1977. Discipline and Punish: The Birth of the Prison. Translated by Alan Sheridan. New York: Pantheon.

Foucault, Michel. 1991. “Governmentality.” In The Foucault Effect: Studies in Governmentality, edited by Graham Burchell, Colin Gordon, and Peter Miller, 87–104. Chicago: University of Chicago Press.

Frankfurt, Harry G. 1971. “Freedom of the Will and the Concept of a Person.” The Journal of Philosophy 68 (1): 5–20.

Harcourt, Bernard E. 2007. Against Prediction: Profiling, Policing, and Punishing in an Actuarial Age. Chicago: University of Chicago Press.

Hart, H. L. A. 1968. Punishment and Responsibility: Essays in the Philosophy of Law. Oxford: Oxford University Press.

Kostka, Genia. 2019. “China’s Social Credit Systems and Public Opinion: Explaining High Levels of Approval.” New Media & Society 21 (7): 1565–1593.

Kundnani, Arun. 2014. The Muslims Are Coming! Islamophobia, Extremism, and the Domestic War on Terror. London: Verso.

Rawls, John. 1993. Political Liberalism. New York: Columbia University Press.

Simester, A. P., and Andreas von Hirsch. 2011. Crimes, Harms, and Wrongs: On the Principles of Criminalisation. Oxford: Hart Publishing.

Taylor, Charles. 1989. Sources of the Self: The Making of the Modern Identity. Cambridge, MA: Harvard University Press.

Velleman, J. David. 1989. Practical Reflection. Princeton: Princeton University Press.

Velleman, J. David. 2000. The Possibility of Practical Reason. Oxford: Oxford University Press.

von Rad, Gerhard. 1972. Genesis: A Commentary. Revised edition. Translated by John H. Marks. Old Testament Library. Philadelphia: Westminster Press.

Wenham, Gordon J. 1987. Genesis 1–15. Word Biblical Commentary, vol. 1. Waco, TX: Word Books.

Westermann, Claus. 1984. Genesis 1–11: A Commentary. Translated by John J. Scullion. Minneapolis: Augsburg Publishing.

Zuboff, Shoshana. 2019. The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. New York: PublicAffairs.