Conviction Under Revisability

The Monk–Prisoner Problem and the Ethics of Irreversible Commitment

Devin C. Arkema

March 2026

Abstract

Irreversible commitment—religious vows, long-term contracts, constitutional entrenchment, AI value-locking—presents a structural ethical puzzle that existing consent theory addresses only partially. Standard accounts of consent focus on the agent's psychological state at the moment of commitment: were they free from coercion, did they understand the terms? This paper argues that psychological voluntariness is necessary but insufficient for ethical legitimacy. What distinguishes legitimate irreversible commitment from structural capture is not the agent's reported experience but the structural conditions of their deliberative architecture.

This paper formalizes that distinction through the Monk–Prisoner Problem: a monk who freely vows permanent enclosure and a prisoner who cannot leave are outwardly identical, yet structurally distinct. The difference lies in revocation authority—who holds the meta-update rule over the closure condition. Four formal criteria are derived from this distinction and applied to four domains where irreversibility raises structural concerns: religious vows, long-term contracts, political constitutions, and AI value-locking. The analysis reveals that structural capture and sovereign surrender are not determined by behavioral observation but by the architecture of the commitment's governance relation.

Keywords: irreversible commitment, revisability, recursive agency, precommitment, structural capture, corrigibility, constitutional entrenchment, consent theory

1. Introduction: The Ethics of Locking In

Among the most enduring puzzles in ethics and political philosophy is the question of when an agent may legitimately bind herself to a course of action in a way that forecloses future revision. This question appears in remarkably diverse domains: the monk who vows permanent enclosure and thereby removes himself from ordinary deliberative life; the contracting party who accepts terms that cannot be rescinded without catastrophic cost; the constitutional framers who entrench certain rights against future amendment; the AI developer who encodes values into a system that will govern its own behavior long after deployment. In each case, an agent freely initiates a structure that constrains future agency—sometimes radically and irreversibly.

The philosophical literature has addressed pieces of this puzzle in isolation. Contract theory has developed doctrines of unconscionability and duress to identify when consent was not freely given. Political philosophy debates the legitimacy of constitutional entrenchment and the extent to which past majorities may bind future ones. Religious ethics explores the nature and limits of vows. AI alignment research grapples with value-locking and the problem of corrigibility under irreversible training. But none of these traditions has produced a unified account of what makes irreversible commitment ethically legitimate in the first place—a formal criterion that could apply across domains and distinguish sovereign self-binding from structural self-capture.

This paper argues that what it terms the Monk–Prisoner Problem provides exactly this unified criterion. The problem is deceptively simple in its statement: a monk who freely vows permanent enclosure and a prisoner who cannot leave are outwardly identical. Both exist within bounded spaces from which they cannot exit. Yet we judge them as morally distinct, and indeed as instantiating structurally different relationships to their own constraint. The question this paper addresses is: what exactly is the structural difference?

The answer proposed here is revocation authority: who holds the meta-update rule over the closure condition? The monk, having exercised intact recursive agency in making the vow, retains—within the structure of the vow itself—the structural power of having been its recursive author. That power does not require that revision remain practically possible in the ordinary sense; it requires that the closure was itself a recursive act, performed under conditions in which revision was structurally available but deliberately foregone. The prisoner has no such authority. The constraint was not authored recursively, and its continuation is not governed by the prisoner's own meta-update rules.

This paper's contribution is threefold. First, it formalizes the Monk–Prisoner distinction using a formal apparatus of recursive agency: Reversibility Symmetry (Rₛ), Recursive Capacity (Cᵣ), and governance continuity. Second, it derives four formal criteria for ethically permissible irreversible commitment that generalize the distinction beyond its original context. Third, it applies these criteria to four contemporary domains—religious vows, contracts, constitutions, and AI value-locking—examining both the criteria's explanatory power and the hard cases at which they break down.

The paper proceeds as follows. Section 2 presents the theoretical framework, developing the concepts of governance continuity, Recursive Capacity, and the deliberative architecture a commitment must pass through to qualify as recursive. Section 3 states the Monk–Prisoner Problem formally and identifies its structural core. Section 4 develops the four criteria for permissible irreversible commitment. Section 5 applies these criteria across the four target domains. Section 6 addresses hard cases—commitments made under partial capture—and Section 7 draws out the implications for philosophy and governance theory.

2. The Theoretical Framework

2.1 Identity as Governance Continuity

The framework developed here begins with a challenge to substance-based accounts of personal identity. If biological organisms persist not through material continuity—cells die and are replaced, proteins degrade and are rebuilt, neural connections are pruned and rewritten—then identity cannot reside in material substance. It must reside elsewhere. The proposal developed in the source texts that ground this paper is that identity persists through governance continuity: the preservation of the governing constraints that regulate a system's behavior across reconstruction. This relationship is formalized as:

S(t+1) = R(S(t), E(t))

Where S is the system state, R is the revision function, and E is the environment. Identity holds when R remains continuous. Identity fails not when the material substrate changes, but when the revision function itself is corrupted, overridden, or abandoned. This analysis extends John Locke's account of personal identity as consisting not in material continuity but in psychological continuity, relocating identity not in consciousness alone but in the persistence of the governing architecture that regulates behavior.

The governance continuity framework identifies four classes of failure modes that threaten the integrity of R: drift (gradual alteration of R without explicit revision), capture (external seizure of the meta-update function), presentism (collapse of R to the immediate state, losing access to historical patterns), and integrity inversion (systematic reversal of R's core commitments). These failure modes are not merely abstract categories; they correspond to recognizable patterns in institutional, personal, and technological governance. The Monk–Prisoner distinction maps onto the capture failure mode with particular precision: structural capture is the condition in which an external party holds the meta-update function, while the agent experiences continuity unaware of the transfer.

2.2 Recursive Capacity and the Indistinguishability Thesis

Recursive Capacity (Cᵣ) is the capacity of an agent to revise the rules governing its own behavior—not merely to update states within a fixed rule-set, but to assess and modify the rule-set itself. Harry Frankfurt identifies this capacity when he distinguishes second-order volitions (desires about desires, or endorsements of one's own motivational states) from mere first-order desires. The present framework generalizes Frankfurt's insight: Recursive Capacity applies to any governance architecture in which the agent can reflect on and revise the rules by which it governs itself, not only to the specific case of volitional endorsement.

The epistemically critical observation is what the framework calls the Indistinguishability Thesis (IT): an agent cannot reliably determine, from within their own deliberative process, whether that process reflects genuine Recursive Capacity or is already shaped by evaluative capture. Deep enough entanglement makes the distinction between authentic deliberation and captured deliberation epistemically unavailable from the inside. An agent who sincerely experiences their deliberation as free may be wrong; the sincerity of the experience is not evidence of its structural independence.

This creates the paper's central diagnostic problem. If behavioral and phenomenal indicators are insufficient to distinguish monk from prisoner, structural indicators are needed. The four criteria in Section 4 are structural precisely because they do not depend on the agent's report of their own deliberative experience. They examine the architecture of the commitment, not the phenomenology. Terminal Recursion names the boundary condition at which Recursive Capacity has collapsed entirely—where the agent is no longer capable of revising the rules governing their own behavior even in principle—and represents the clearest case of Prisoner-type commitment.

2.3 Reversibility Symmetry and the Commitment Architecture

Reversibility Symmetry (Rₛ) quantifies the total cost of exiting a commitment across all relevant dimensions: financial, psychological, social, informational, and identity-constitutive. The key insight is not the formula itself but what it reveals about commitment structures: reversal costs can compound over time in ways that were not fully foreseeable at entry, and they compound unevenly across dimensions. Identity-constitutive costs—the cost of revising not just a position but who one takes oneself to be—are frequently the most invisible at entry and the most rapidly compounding over time. A commitment that appeared reversible at closure may become practically irreversible as these costs accumulate.

Any genuinely recursive commitment must pass through four structural stages that constitute what this paper calls the Ethical Loop: Free Won't (the genuine capacity to decline the commitment, not merely the absence of physical coercion); Recursion (review of historical patterns and forward conditions, not merely assessment of the immediate moment); Witness (external verification that the deliberative process was not captured, independent of the agent's own testimony); and Interruption consideration (awareness of the conditions under which the commitment would need to be revisited, specified in advance rather than at the moment of crisis). Together these stages constitute the commitment architecture whose preservation the four criteria in Section 4 will test. A commitment that passes through all four stages has the structural signature of recursive authorship; a commitment that bypasses any stage raises the possibility of submission rather than sovereign choice.

3. The Monk–Prisoner Problem

With the theoretical apparatus in place, the Monk–Prisoner Problem can be stated precisely. Consider two individuals:

(1) M: a monk who has freely vowed permanent enclosure in a contemplative religious order, after an extended period of novitiate formation, with full knowledge of the conditions of enclosure, before witnesses who could attest to the recursive character of his deliberation, and with awareness that his tradition retains institutional authority to grant dispensation under extraordinary circumstances.

(2) P: a prisoner who has been confined without choice, without knowledge of the duration of confinement, without any structural mechanism by which the confinement could be revised or contested, and through a process that never invoked anything resembling deliberate recursive authorship.

M and P are outwardly identical: both inhabit bounded spaces from which they cannot exit. Their behavioral repertoires are similarly constrained. Their phenomenal experience of constraint may, in some respects, resemble each other. Yet we judge them as morally distinct, and our intuitions about this distinction are robust across traditions.

The Monk–Prisoner Problem is the philosophical challenge of making this intuitive distinction theoretically rigorous. Several candidate answers in the existing literature are inadequate. The behavioral answer—that M chose to enter while P did not—fails because choice in the relevant sense requires unpacking: what conditions must obtain for an act of choosing to constitute genuine recursive authorship rather than submission? The phenomenological answer—that M experiences his enclosure as meaningful while P may experience his as suffering—fails because phenomenal experience is precisely what the Indistinguishability Thesis places under suspicion: a structurally captured agent may experience their capture as authentic and meaningful. The consent answer—that M consented while P did not—is the closest to correct but requires formal specification: consent at what point in time, under what structural conditions, with what degree of understanding of the commitment's reversal costs?

This paper's answer is that M and P differ in four structural conditions that jointly constitute legitimate irreversible commitment. M satisfies all four; P satisfies none. The interesting philosophical and legal cases are those in which the conditions are partially met—and it is this partiality that the criteria developed in Section 4 are designed to diagnose.

The formal conditions that distinguish M from P are:

(1) M possessed intact Recursive Capacity at the time of closure; P's deliberative architecture, if any, was not engaged in a way that preserved recursive agency.

(2) M had adequate knowledge of the reversal costs of enclosure; P either had no such knowledge or was unable to act on it.

(3) M holds, as the recursive author of the vow, a structural relationship to the closure condition that P lacks; the institutional tradition's retention of dispensation authority preserves a form of meta-update authority even within the permanent vow's structure.

(4) M's commitment was reached through a process that respected all four stages of the Ethical Loop; P's confinement did not pass through any analogous deliberative architecture.

Jon Elster's analysis of precommitment as voluntary self-binding—the Ulysses who ties himself to the mast to resist the Sirens—illuminates part of the monk's situation but does not resolve the problem. Elster's account treats precommitment as a second-order rational strategy: binding the future self to prevent akratic deviation. But this analysis does not distinguish between precommitment structures that preserve the agent's meta-update authority and those that eliminate it. Ulysses precommits but instructs his crew not to untie him regardless of his future commands; the precommitment's authority derives from Ulysses' explicit instruction at a moment of clear-sightedness. The monk's commitment is structurally similar in this respect: its authority derives from the recursive character of the act that produced it. But the analysis must extend to cases—more common in law and ethics than in mythology—where the recursive character of the closure act is precisely what is in question.

4. Four Formal Criteria for Permissible Irreversible Commitment

4.1 The Recursive Standing Criterion (RSC)

The committing agent must possess intact Recursive Capacity at the time of closure. An agent whose Recursive Capacity is compromised—through evaluative capture, informational deprivation, coercive pressure, or conditions that have collapsed the meta-level of their governance architecture—cannot be a recursive author of the closure. The commitment they make is not sovereign surrender but structural capture, even if no specific act of coercion can be identified.

The RSC is more demanding than standard consent requirements in legal and philosophical doctrine, which typically require only the absence of explicit coercion and some threshold level of understanding. The RSC requires the positive presence of recursive agency: the agent must have been capable of genuine meta-level assessment of the commitment before making it. This means examining not only the moment of closure but the deliberative process leading to it—whether the conditions leading to closure preserved or degraded the agent's capacity for recursive assessment.

Philip Pettit's republican account of freedom as non-domination is useful here. Pettit argues that freedom requires not merely the absence of interference but the absence of a structural condition under which interference would be possible without the agent's effective resistance. On this view, an agent's freedom is compromised not only when they are actually interfered with but when they exist in a social relation that makes arbitrary interference a live possibility. The RSC applies an analogous structural analysis to the deliberative architecture of commitment: the relevant question is not whether any specific act of coercion occurred but whether the governance structure leading to closure was of a kind that preserved or compromised the conditions for genuine recursive agency. Minimum Exit Viability—the minimum intact recursive capacity required for a commitment to register as recursively authored—sets the floor below which closure produces Prisoner-type outcomes regardless of surface voluntariness.

4.2 The Bounded Reversibility Criterion (BRC)

The agent must have had adequate knowledge of the commitment's reversal costs at the time of closure, and those costs must have been bounded—not growing without limit as a result of structural features of the commitment itself. A commitment whose reversal costs were concealed at entry, or whose design produces compounding exit costs without bound, fails the BRC regardless of the agent's subjective experience of voluntariness.

The BRC has a specific relationship to existing economic analyses of commitment. Richard Posner's economic analysis of contract law suggests that binding commitments are efficiency-enhancing precisely because they signal credibility: a commitment the agent cannot easily abandon is more valuable to the promisee than one they can exit costlessly. On this account, high reversal costs are desirable features of commitment structures, not failures. The BRC does not dispute this but distinguishes between two types of reversal costs that the efficiency argument conflates. Constitutive reversal costs are inherent to the commitment type: a monk who vows enclosure necessarily incurs the cost of physical limitation; an employee who accepts a non-compete necessarily incurs some cost to future employment mobility. These costs are part of what the commitment means, were foreseeable in principle at closure, and their presence does not violate the BRC. Compound reversal costs, by contrast, accrue through structural features of the commitment that were not adequately disclosed or foreseeable at entry: the rapid compounding of identity-constitutive costs as the agent's self-conception becomes entangled with the institution, social network effects that make exit catastrophic rather than merely costly, informational asymmetries that grow as institutional-specific knowledge is acquired. Commitment structures that produce compound reversal costs systematically violate the BRC even when constitutive costs were within the bounds of informed consent.

This distinction has direct implications for how unconscionability doctrine should analyze long-term commitments. Current doctrine addresses substantive unconscionability—whether the terms are unreasonably one-sided—but does not clearly distinguish between the one-sidedness that is constitutive of the commitment type and the one-sidedness that compounds asymmetrically through structural design. The BRC provides the theoretical basis for this distinction.

4.3 The Meta-Update Authority Criterion (MUAC)

The committing agent must, at the time of closure, be the recursive author of the closure condition: the rule by which the commitment governs future behavior must derive its authority from their recursive act of committing, not from external imposition. Furthermore, the governance structure of the commitment must preserve some form of meta-update authority over the closure condition—whether held by the agent, by a trusted institutional proxy, or by a structured collective process.

The MUAC is not a requirement that revision remain easy or even practically likely. It is a requirement that the meta-update rule Rₐ over the commitment remains operative somewhere in the governance architecture. A monk's permanent vow satisfies the MUAC when the institutional tradition retains dispensation authority as a structural feature of the vow's governance—Rₐ is held in trust at the institutional level, available for invocation under conditions specified by the tradition. A prisoner's confinement fails the MUAC entirely: the closure condition is not governed by any rule the prisoner authors or can invoke. A constitutional entrenchment satisfies the MUAC when demanding-but-accessible amendment procedures preserve the collective meta-update function; it fails when provisions are formally or functionally unamendable.

An important objection to the MUAC challenges its treatment of one-time authorship. If an agent recursively authors a closure in a single act—setting a rule that future behavior must conform to and then closing off revision—has the MUAC been satisfied? The objection is that this looks like legitimate recursive closure: the agent used their Recursive Capacity, they held Rₐ at the time of closure, and the commitment's permanence is just what the commitment means. The MUAC's response is that one-time authorship carries governing authority only when the closure was itself recursively verified—when the act of committing was not merely performed by an agent with Recursive Capacity but was externally witnessed as such. This is the dependency that connects MUAC to the ELC, discussed below.

The self-nullification objection raises a closely related challenge: if an agent uses their Recursive Capacity to commit to never revising a certain class of commitments, have they legitimately self-nullified the very Recursive Capacity that the RSC and MUAC require? This objection has genuine force. The framework's response is that legitimate self-nullification is possible—the monk case is precisely this—but requires all four criteria to be satisfied at the moment of self-nullification. An agent can recursively authorize a commitment to non-revision only if, at the time of that authorization, they possessed intact Recursive Capacity (RSC), had adequate knowledge of reversal costs (BRC), were the recursive authors of the closure (MUAC), and reached the commitment through an Ethical Loop process (ELC). Self-nullification that bypasses any of these conditions does not constitute legitimate recursive closure but structural capture initiated from the inside.

4.4 The Ethical Loop Criterion (ELC)

The commitment must have been reached through a process that respects the full Ethical Loop: Free Won't (genuine capacity to decline), Recursion (review of history and forward conditions), Witness (external verification), and Interruption consideration (awareness of conditions under which the commitment would need to be revisited). A commitment reached through the short-circuiting of any of these stages is not a recursive act but a submission. The ELC does not require that all four stages be formally executed; it requires that the governance structure that produced the commitment preserved the structural conditions for each stage.

A commitment made impulsively, under social pressure that foreclosed genuine deliberation, without consideration of historical patterns or future conditions, and without external verification, fails the ELC even if the agent experienced it as voluntary. The Indistinguishability Thesis is most acutely operative at precisely the moments when the ELC is most needed: high-stakes, emotionally charged, identity-constitutive commitments are exactly those in which an agent is most likely to be in a state of evaluative capture without being able to detect it. The ELC's Witness stage is structurally necessary because it provides external verification that the Indistinguishability Thesis renders impossible from inside.

The Conviction–Revisability Distinction is relevant here: conviction that has survived structured challenge is legitimate; conviction that has avoided structured challenge is suspect. The ELC requires that irreversible commitment be treated as a high-stakes position that demands the highest level of external review, because the Indistinguishability Thesis applies most acutely precisely to commitments of great personal significance.

4.5 The Relationship Between MUAC and ELC

The relationship between MUAC and ELC requires explicit clarification to prevent criterion bleed. Both concern the governance of the closure act, but they operate at different levels and neither subsumes the other. MUAC is locational and authority-based: it identifies who holds the meta-update rule Rₐ over the commitment. ELC is processual and evidential: it establishes whether the exercise of that authority was reached through a process that preserved the Ethical Loop's four stages.

They connect in this way: MUAC establishes the authority of the closure (the commitment is legitimated by the fact that the agent held and exercised Rₐ); ELC provides the evidentiary basis for that authority claim (the exercise of Rₐ was genuinely recursive only if all four stages were operative). The ELC is thus the process criterion that authenticates the exercise of MUAC—not a parallel criterion testing a separate dimension, but the evidential structure that gives MUAC's authority claim its substance.

A commitment where MUAC is satisfied but ELC fails—the agent holds Rₐ but exercised it through a compromised process—has formal authority without substantive legitimacy, analogous to a judicial decision issued by a court with proper jurisdiction but through a fatally defective procedure. A commitment where ELC is satisfied but MUAC fails—the process was genuinely deliberative but Rₐ was then structurally transferred away—has substantive legitimacy without formal authority, analogous to a valid democratic election whose winner is stripped of executive power. The four criteria thus form two pairs: RSC and BRC address the conditions of the committing agent at closure; MUAC and ELC address the authority and authenticity of the closure act itself.

Together, the four criteria (RSC, BRC, MUAC, ELC) provide a formal account of the conditions under which irreversible commitment constitutes sovereign surrender rather than structural capture. In the clear monk case, all four conditions are met. In the clear prisoner case, all four conditions fail. In the hard cases that interest both philosophy and law, the conditions are partially met—and it is this partiality that the four criteria are designed to diagnose.

5. Application: Four Domains of Irreversible Commitment

5.1 Religious Vows

Religious traditions have long grappled with the ethics of permanent vows. The Catholic tradition's treatment of solemn vows—particularly the vow of chastity and the vow of enclosure in contemplative orders—provides the clearest analog to the Monk case. Church law treats solemn vows as permanently binding, though it retains institutional authority to dispense from certain vows under specific conditions. This institutional retention of dispensation authority is, in the formal framework developed here, a structural preservation of meta-update authority: the Church, as the institutional context of the vow, holds Rₐ in trust even when the individual vow-maker cannot invoke it.

Applying the four criteria to religious vows reveals both their justificatory structure and their pressure points. The RSC requires that the novice entering a permanent vow possesses intact Recursive Capacity—genuinely capable of declining, not yet captured by institutional identity or social pressure that forecloses genuine deliberation. Traditional religious formation processes—the extended novitiate period, the graduated stages of commitment before final vows—can be understood as structural attempts to satisfy this criterion: creating deliberative conditions in which recursive standing is cultivated and tested before irreversible closure. The length and intensity of formation, read through the framework, is governance preparation rather than primarily spiritual preparation—ensuring that the RSC and ELC conditions are met before the irrevocable step.

The BRC is more problematic. Reversal costs in religious enclosure are not merely logistical but deeply psychological, social, and identity-constitutive. A monk who has lived within an order for years before final vows faces reversal costs that were structurally predictable but not fully experientially available at the time of closure. The BRC distinguishes between constitutive reversal costs—inherent to the vow's religious significance—and compound costs that accrue asymmetrically beyond what was foreseeable. Some of the reversal costs of enclosure belong to the first category; others accumulate in ways that traditional formation processes address unevenly.

The MUAC is where the framework reaches the boundary of its jurisdiction over religious vows, and it is important to be precise about that boundary. Catholic canon law's dispensation mechanism preserves a form of institutional meta-update authority: Rₐ is held, in trust, by a governance structure above the individual. But this reading depends on contested theological and institutional assumptions that the framework cannot adjudicate from outside. The philosophically precise question for any religious tradition is not whether permanent enclosure is legitimate, but who holds Rₐ over vow interpretation and dispensation, and whether that authority is itself recursively accountable.

The hardest challenge from within the religious tradition concerns whether the RSC's temporal framing adequately handles genuine conversion. A monk who has undergone authentic spiritual formation may have become the kind of person for whom the vow no longer registers as constraint—who would, if the vow were lifted, maintain its terms by choice. The framework might seem to misdiagnose this transformation: the monk's preferences have shifted to align with the institution's requirements, which structurally resembles evaluative capture. The response is that the RSC's criterion is deliberative, not preferential: what matters is whether Recursive Capacity was intact at the time of closure, not whether the agent currently wants to remain. Authentic transformation is downstream of recursive commitment; its authenticity consists partly in the fact that it was not imposed but freely entered. The Indistinguishability Thesis, however, generates residual uncertainty here: if capture and conversion can produce similar phenomenal states, the Witness criterion's external verification becomes especially important in religious formation contexts.

5.2 Long-Term Contracts and Unconscionability

Contract law has developed the doctrine of unconscionability precisely to address commitments that appear formally voluntary but are substantively extractive. An unconscionable contract is typically characterized by procedural unconscionability (unfairness in the formation process: absence of meaningful choice, unequal bargaining power, failure to understand terms) and substantive unconscionability (unfairness in the terms themselves: severe imbalance in obligations or benefits).

The formal analysis of contracts cuts across the two major theoretical accounts of contractual obligation. Charles Fried's promissory account grounds contractual obligation in the moral weight of a promise: breaking a contract wrongs the promisee by violating the trust that gave the promise its social meaning. Patrick Atiyah's reliance account grounds obligation in the harm caused by disappointed reliance rather than in the act of promising itself. The four-criteria framework does not adjudicate between these accounts but locates the prior question both presuppose: under what structural conditions is the act of commitment—whether understood as a promise or as inducement to reliance—recursively authorized in the first place?

The framework maps onto unconscionability doctrine with considerable precision. Procedural unconscionability corresponds to failure of the RSC and ELC: the weaker party lacked intact Recursive Capacity or the commitment was not reached through a process resembling the Ethical Loop. Substantive unconscionability corresponds to failure of the BRC: the terms create an asymmetric reversal cost structure that grows without bound. But the analysis extends beyond existing unconscionability doctrine in two directions. First, it reveals that procedural unconscionability can occur without overt coercion—through the structural degradation of Recursive Capacity described by the Indistinguishability Thesis. A contracting party subjected to prolonged negotiation, manufactured urgency, and information overload may reach a point at which Recursive Capacity is functionally compromised without any single identifiable act of coercion. Second, the MUAC raises a question that unconscionability doctrine does not clearly address: even when a contract was voluntarily entered and the terms were understood, do the terms themselves respect the committing party's meta-update authority? Certain long-term contracts can be structured to eliminate the practical capacity for revision in ways that are not identifiable as unconscionable under current doctrine but that violate the MUAC.

The promissory account presents the sharpest challenge to the structural analysis. On Fried's account, contractual obligation derives from the moral weight of the promise itself—from the trust created in the promisee by the act of promising—rather than from the structural conditions of the promisor's deliberation. If Fried is right, a promise may generate binding obligation even if the RSC was compromised, because the obligation runs through the promisee's reasonable expectations, not the promisor's internal governance state. The structural account's reply is that MUAC forces a distinction Fried's account does not make: a promise that eliminates the promisor's capacity to revise their governance position creates an obligation structure that, over time, becomes indistinguishable from structural capture regardless of its initial promissory validity. The promissory account identifies what creates the prima facie obligation; the four criteria identify whether that obligation remains ethically sustainable as conditions change. A contract that was both well-made and whose terms preserve the promisor's meta-update authority satisfies both frameworks; it is the cases where these come apart that the structural analysis is designed to address.

5.3 Constitutional Entrenchment

Constitutional entrenchment—the practice of placing certain constitutional provisions beyond the reach of ordinary amendment procedures—raises the Monk–Prisoner problem at the political scale. Constitutions entrench provisions because their framers believe those provisions' permanence is itself valuable: they provide stability, protect against majoritarian overreach, and express long-term commitments that should not be subject to short-term political fluctuation. But entrenchment forecloses future democratic revision in ways that raise deep questions about self-governance across generations.

The classic objection—voiced in various forms by Jefferson, Paine, and more recently by constitutional scholars such as Akhil Reed Amar—is that each generation must have the right to govern itself, and that binding future generations to the decisions of past majorities is a form of domination. The standard response invokes consent by acquiescence (living under a constitution implies tacit consent to its authority) or value-preservation (certain rights are so fundamental that their protection outweighs generational autonomy).

The MUAC reframes this debate. The central question is whether current and future citizens hold Rₐ—the meta-update rule—over the entrenched provisions. If they do not—if entrenched provisions are genuinely unamendable by any existing process—then constitutional entrenchment violates MUAC regardless of the historical voluntariness of adoption. If they do—if the constitutional order retains structural mechanisms for revisiting entrenched provisions, even with extraordinarily demanding thresholds—then entrenchment may satisfy MUAC provided the other criteria are met.

Before applying the MUAC analysis to eternity clauses, the strongest defenses of constitutional unamendability deserve direct engagement. Two are most powerful. The integrity argument, developed most rigorously in Ronald Dworkin's account of the constitution as integrity, holds that certain provisions deserve permanent protection because they represent the community's best moral reasoning at its most reflective—not its most majoritarian or most expedient. Rights are entrenched precisely because democratic majorities have a structural tendency to shortcut deliberation when political pressure is highest. The stability argument, associated with theorists from Jon Elster to Frederick Schauer, holds that constitutional permanence provides a coordination focal point and credible precommitment device: a constitution that can be easily dismantled offers insufficient protection against democratic backsliding or executive overreach, and the value of the precommitment is in part a function of its unconditional character.

Both arguments are serious. The response to the integrity argument is that Dworkin's account of constitutional rights as representing the community's highest-quality moral reasoning actually undermines eternity clauses rather than supporting them. If rights are entrenched because they represent the community's best moral reasoning, they should remain revisable when that reasoning has advanced—provided revision meets appropriately demanding conditions. Eternity clauses implicitly assert the permanent superiority of the founding moment's moral reasoning over all future reasoning, a claim that integrity theory does not support.

The response to the stability argument is structural. A provision requiring a ninety-percent legislative supermajority, ratification by three-quarters of constituent units, and a ten-year waiting period satisfies both the stability argument's precommitment demand and the anti-majoritarian concern—while preserving MUAC at the collective level. The stability argument establishes that high amendment thresholds serve legitimate purposes; it does not establish that eliminating meta-update authority entirely is necessary or proportionate to those purposes.

The MUAC argument against eternity clauses faces a further challenge from within the stability framework: if meta-update authority is formally preserved but procedurally inaccessible—requiring unanimous agreement that no realistic coalition could achieve—has MUAC been satisfied? The stability theorist can argue that procedural inaccessibility is not the same as formal elimination; Rₐ exists in principle even if it cannot be invoked in practice. The framework's response is that MUAC requires not merely formal preservation of Rₐ but functional accessibility: a revision pathway so demanding that no realistic process can traverse it is structurally equivalent to no pathway at all, because the governance consequence is identical—the closure condition cannot be revised. This functional accessibility standard is more demanding than the stability argument's formal preservation standard, but it corresponds to what legitimate self-governance requires: the meta-update rule must be available as a live option, not merely a theoretical possibility.

5.4 AI Value-Locking and Corrigibility

The question of irreversible commitment in AI systems is among the most practically urgent applications of the Monk–Prisoner distinction. AI alignment research has long grappled with the tension between value stability (ensuring that AI systems maintain safe values over time) and corrigibility (ensuring that AI systems remain revisable by human overseers). Value-locking—the embedding of values in AI systems in ways that cannot be modified after deployment—is the extreme form of this tension: it treats the commitment to safety as requiring irreversible closure.

From the perspective of the four-criteria framework, the key question is whether value-locking in AI systems constitutes Monk-type or Prisoner-type commitment. The problem of epistemic hardening is particularly salient here: the tendency of prior safety judgments to acquire illegitimate present authority, converting provisional assurance into institutional inertia. Epistemic hardening is a form of the capture failure mode at the institutional level—the governance system loses access to its own capacity for revision and collapses into defending what it previously authorized.

Value-locking in AI systems raises all four criteria simultaneously. The MUAC question is most directly answerable: a corrigible AI system—one that accepts correction, modification, and value revision from human overseers—satisfies MUAC to the extent that the overseers hold Rₐ over the AI's value commitments. An AI system from which all corrigibility has been removed fails MUAC regardless of how carefully its values were selected, because no meta-update rule remains operative. The concept of distributed corrigibility—corrigibility understood as a governance property of the entire sociotechnical stack rather than a property of the model alone—can be understood as an institutional answer to the MUAC requirement: ensuring that Rₐ is held at the system level even when individual model components cannot revise their own values.

The ELC analysis has particularly sharp implications. The Ethical Loop requires that irreversible commitment be reached through a process that preserves Free Won't, Recursion, Witness, and Interruption. AI deployment decisions are rarely reached through processes that satisfy these conditions: commercial pressures foreclose deliberation, competitive dynamics compress the Recursion stage, Witness is often limited to internal review teams, and Interruption mechanisms are rarely pre-committed with sufficient specificity. The concept of a living safety case—a safety case whose legitimacy depends not merely on the quality of its initial argument but on explicit criteria for reassessment, downgrade, rollback, and retirement—is precisely an institutional attempt to satisfy the ELC requirement: embedding the Ethical Loop into the governance architecture of AI deployment.

The sharpest challenge from within AI alignment research is that the RSC may not apply symmetrically to current AI systems. If a current AI system lacks Recursive Capacity at the meta-level—if it can update states but cannot revise the rules governing those updates—then value-locking does not override an existing Recursive Capacity; it constrains a system that never possessed one. The monk analogy breaks down: we would not say a chess program is imprisoned by its rules in any ethically relevant sense. The framework's response is twofold. First, this breaks the analogy with current systems while preserving it for systems that develop genuine meta-level recursive capacity—and the framework should be applied at whichever level of recursion the system actually achieves. Second, this shifts the ethical locus from the AI to the human governance structure surrounding it. The question of whether value-locking is Prisoner-type becomes a question about the designers' and operators' MUAC: do they hold Rₐ over the AI's value commitments? If the deployment conditions eliminate the human governance system's ability to revise those values even when urgent reasons emerge, the failure is not the AI's captivity but the human governance system's self-imposed closure. Corrigibility, on this account, is about preserving the human meta-update function, not about respecting AI agency that may not yet exist at the required level.

6. The Partial Capture Problem: Graduated Irreversibility

The four criteria, applied to clear cases, yield clear verdicts. The monk satisfies all four; the prisoner satisfies none. But the philosophically and practically important cases are the graduated ones, in which the criteria are partially satisfied.

6.1 The Compounding Problem

The most common form of graduated irreversibility involves the compounding of reversal costs over time in ways that were not fully anticipated at closure. A person who accepts employment under a non-compete agreement understands, at closure, that they cannot work for competitors for a defined period. What they may not fully anticipate is the compounding of identity reversal costs as their professional identity becomes entangled with the employer's brand, the information reversal costs as their expertise becomes domain-specific in ways that reduce employability elsewhere, and the network reversal costs as their professional relationships are mediated through the employer's infrastructure. At closure, Rₛ was finite; a year later, it may be substantially higher.

The framework's analysis reveals that this compounding typically occurs unevenly across cost components. Financial reversal costs may be calculable and known at closure; psychological, social, and identity reversal costs compound in ways that are both less visible and more rapid. Identity reversal cost can become the most expensive component—it requires not just changing a position but changing who one takes oneself to be. This cost is most likely to be invisible to the agent experiencing it, because identity operates below the level of explicit accounting. The BRC's demand for bounded reversal costs is most frequently violated through this identity-constitutive compounding, not through financial extraction.

6.2 Social Pressure and Recursive Standing

A second form of graduated irreversibility involves the gradual erosion of Recursive Capacity through social pressure that did not, at any single moment, constitute coercive foreclosure of choice but accumulated into evaluative capture. The pattern involves social systems that don't announce their rules but allow agents to feel them—through informal norms, through the structure of belonging, through the way identity becomes entangled with the institution. In each case, no individual instance of social pressure constitutes coercion, but the cumulative process produces a state of evaluative capture: the agent's preferences have been shaped to align with the system's requirements in ways that make genuine deliberative refusal practically unavailable.

A commitment made in this state fails the RSC—not because any specific act of coercion occurred, but because the cumulative process of social shaping has degraded Recursive Capacity to the point where the commitment is not genuinely recursive. This analysis has direct implications for contracts signed under informational asymmetry, vows made under social pressure, and AI systems trained on data that embeds evaluative capture. In each case, the question is not whether there was a specific identifiable act of coercion but whether Recursive Capacity was intact at the time of closure. The latter is a structural question that requires examining the process leading to closure, not merely the act of closure itself.

6.3 The Limiting Case: Nominal MUAC

The hardest case for the four-criteria analysis is not the case where all criteria obviously fail but the one where three appear satisfied and the fourth fails in a way that is systematically obscured from the committing party. Consider the governance structure of a multi-level marketing compensation arrangement. Entry appears voluntary: the RSC appears satisfied, because the participant had the capacity to decline and did not face any single identifiable act of coercion. Reversal costs appear known: the BRC appears at least partially satisfied, because the participant was told, in general terms, that earnings depend on recruiting and that exit involves losing invested capital. The loop criterion appears met: the ELC appears partially satisfied, because the participant attended information sessions, consulted with others, and made what felt like a considered decision. Surface inspection returns a Monk-type verdict.

But the MUAC analysis reveals a structural failure invisible to these criteria. The meta-update rule governing the participant's continued participation is held not by the participant but by the organization's compensation architecture. The participant cannot revise their position in the network without dismantling the relationships through which their income flows, and the organization controls the terms of those relationships unilaterally. Exit requires not merely leaving but destroying the network the participant spent years building—a form of identity and social reversal cost that compounds on an accelerating trajectory the participant was never in a position to price at entry.

More precisely: the information sessions at which the ELC appeared to be satisfied were conducted by parties whose economic interest was in the participant's continued participation. The Witness layer, formally present, was structurally captured. External verification was available in form but not in substance. This case matters because it shows that voluntariness and MUAC can come apart precisely in the cases the framework is designed to diagnose. The participant was genuinely free in the moment of choice, but was choosing within a governance structure never designed to allow meta-update authority to be theirs. It is a case a referee might describe as: looks Monk-type in the courtroom, is Prisoner-type in the architecture.

6.4 The Graduated Legitimacy Spectrum

Rather than treating commitment as binary—either legitimate or not—the four-criteria framework suggests a graduated legitimacy spectrum based on the degree to which each criterion is satisfied. Full Monk-type commitment (all four criteria fully satisfied) represents one end; full Prisoner-type commitment (all four criteria failed) represents the other. Between these poles is a range of partially legitimate commitments in which some criteria are fully satisfied and others are partially or fully failed.

This graduated approach has practical implications. A commitment in which the RSC is fully satisfied but the BRC is partially violated—because reversal costs were known but significantly underestimated—is partially legitimate. The agent's commitment retains the authority of recursive authorship but was made under conditions of informational insufficiency. Remedies for such commitments need not treat them as void ab initio but can address the specific dimension of failure: providing better information about reversal costs, creating mechanisms for cost adjustment as compounding occurs, or establishing structural checkpoints at which the commitment can be reviewed.

This graduated approach is more consistent with the intuitions underlying existing legal doctrine than a binary legitimate/illegitimate distinction. Courts applying unconscionability doctrine typically do not void contracts entirely but sever unconscionable terms, adjust obligations, or craft remedies proportionate to the degree of unfairness. The framework provides the theoretical basis for this proportionality: the appropriate remedy for partial failure of the criteria is proportionate to the degree of failure, addressing the specific dimension of structural compromise rather than treating the entire commitment as nullified.

7. Discussion: Implications for Philosophy and Governance

7.1 Reframing Consent Theory

The most fundamental implication of the Monk–Prisoner analysis is a reframing of consent theory. Standard accounts of consent—in both legal and philosophical traditions—focus on the agent's state of mind at the time of commitment: did they understand what they were agreeing to, and were they free from coercion? The framework shifts focus from the agent's psychological state to the structural conditions of the deliberative process. The question is not whether the agent felt free but whether the agent's governance function remained intact.

This shift has three consequences. First, it captures cases of structural capture that standard consent theory misses. The MLM case in Section 6.3 illustrates the gap: three of the four criteria appear satisfied on standard consent grounds, but the MUAC exposes an architectural failure invisible to psychological assessment. The participant was genuinely free in the moment of choice but was choosing within a governance structure designed to prevent meta-update authority from ever being theirs. Standard consent theory's focus on the moment of agreement cannot detect this, because the failure is not in the moment but in the architecture surrounding it. The Indistinguishability Thesis explains why the agent cannot detect it from inside either.

Second, the structural focus provides a more principled account of why certain categories of consent are presumptively suspect: consent obtained under conditions of institutional power asymmetry (employment, medical, carceral contexts), consent obtained under time pressure or manufactured urgency, consent to complex terms under information asymmetry. These are conditions that predictably degrade Recursive Capacity or compromise the independence of the Witness layer without constituting the kind of discrete coercive act that standard duress doctrine addresses. The structural analysis does not require finding the coercive act; it requires asking whether the deliberative architecture was intact.

Third, the structural focus redirects institutional design toward the preservation of deliberative conditions rather than the documentation of deliberative outcomes. Standard consent frameworks focus on obtaining evidence of consent—signatures, acknowledgments, witnessed declarations. The structural approach focuses on the conditions that make consent recursively authoritative: processes that preserve Free Won't, support Recursion, maintain genuinely independent Witness, and include pre-committed Interruption thresholds. A commitment backed by extensive documentation but reached through a compromised deliberative process is no more legitimate than one backed by no documentation at all.

7.2 Constitutional Theory and the Persistence of Authority

The application of the Monk–Prisoner analysis to constitutional entrenchment contributes to ongoing debates about the sources of constitutional authority. The standard intergenerational consent debate asks whether past constitutional choices can legitimately bind future generations. The MUAC provides a more precise version of this question: do future citizens hold the meta-update rule over entrenched constitutional provisions?

The answer depends on the constitutional architecture. A constitution that provides amendment procedures for all provisions—however demanding the threshold—preserves MUAC at the collective level. A constitution that designates certain provisions as unamendable fails MUAC for those provisions. The unamendable provisions are Prisoner-type commitments at the political scale: the original framers authored the closure, but no living or future generation can invoke Rₐ over it. This analysis yields a substantive constitutional design principle: eternity clauses are structurally indefensible regardless of their substantive content. The appropriate response to concerns about rights protection is not to make rights constitutionally unamendable but to make the amendment procedure sufficiently demanding that amendment requires the full exercise of collective recursive agency through the Ethical Loop.

7.3 AI Alignment and the Architecture of Corrigibility

The AI alignment application connects directly to the emerging literature on corrigibility as a distributed governance property. Corrigibility must be treated not solely as a model property—a feature of the AI system's internal architecture—but as a governance property of the entire sociotechnical stack. This claim is underwritten by the MUAC: what matters is not whether the AI system itself retains Rₐ over its own values, but whether the governance system surrounding it does.

The Monk–Prisoner framework makes a stronger claim: the distribution of meta-update authority is not merely a governance design choice but an ethical requirement for the legitimacy of AI value-commitments. An AI system whose values are permanently locked—regardless of how carefully those values were selected—is in a Prisoner-type relationship to those values when no meta-update rule remains operative over the closure condition. A safety case that cannot lose standing is not a safety case; it is an eternity clause.

7.4 Toward a Structural Ethics of Commitment

The four criteria, the graduated legitimacy spectrum, and the four domain applications together point toward what might be called a structural ethics of commitment: an approach to evaluating the ethical status of commitments that focuses on the structural conditions of their formation, maintenance, and revision rather than on the psychological experience of the committing agent.

The relation of this structural ethics to the existing literature on precommitment and self-binding is worth making explicit. Jon Elster's tradition treats voluntary self-binding as a subset of rational choice, asking when it is rational to constrain one's future self. Harry Frankfurt's tradition treats it as a question about the structure of the will. Contract theory's major traditions ask whether the commitment was promised or relied upon and whether the resulting obligation is efficient or unjust. The structural ethics proposed here is not a competitor to these traditions but a deeper layer beneath them: it identifies the conditions under which the acts of rational self-binding, second-order endorsement, promising, reliance-inducing, and contracting are themselves recursively authorized. Without those conditions, the traditions' more specific evaluations are building on sand. With them, the traditions acquire a common foundation.

What the present framework contributes is the formal apparatus needed to make the deepest intuitions of these traditions precise, testable, and institutionally actionable—replacing the current patchwork of domain-specific doctrines with a unified structural account. The intuition that consent requires genuine deliberative freedom, not merely the absence of explicit coercion; the intuition that obligations which cannot be revisited eventually become conscription; the intuition that institutional governance retains legitimacy only as long as it preserves the conditions for its own revision—these intuitions are what the four criteria formalize.

8. Conclusion

The Monk–Prisoner Problem is not merely an interesting philosophical puzzle. It is the central diagnostic for a class of ethical failures that existing theory has struggled to address: commitments that are nominally voluntary but structurally self-undermining; closure conditions that appear consensual but violate the structural requirements of recursive agency; governance architectures that lock in values or obligations in ways that eliminate the meta-update authority that legitimate commitment requires.

This paper has formalized the Monk–Prisoner distinction, derived four formal criteria for ethically permissible irreversible commitment (RSC, BRC, MUAC, ELC), and applied them to four contemporary domains: religious vows, contracts, constitutions, and AI value-locking. Each application reveals both the criteria's explanatory power and the distinctive pressure points at which irreversibility becomes structurally problematic. The paper has also addressed the hard cases of graduated irreversibility—commitments made under partial capture—and proposed a graduated legitimacy spectrum in which the appropriate remedy is proportionate to the specific dimension of structural compromise.

Three broader conclusions follow from this analysis. First, voluntariness is not sufficient for legitimate irreversible commitment. What is required is recursive authorship under conditions of intact Recursive Capacity, bounded and known reversal costs, preserved meta-update authority, and an Ethical Loop in the formation process. Second, the proper institutional response to concerns about stability and permanence is not to eliminate meta-update authority but to make its exercise sufficiently demanding—sufficiently deliberative in structure—that it can only be invoked through the full architecture of recursive agency. A commitment structure that makes revision maximally difficult while preserving a live revision pathway is Monk-type; one that eliminates the pathway entirely is Prisoner-type, and no legitimate stability interest requires the second when the first is available.

Third, and most fundamentally, the ethics of irreversible commitment is not primarily a question about psychology—what the agent felt or intended—but about architecture: what governance structures were operative at the time of closure and what governance structures remain operative afterward. The difference between the monk and the prisoner is not outward behavior but the governance relation to their own constraint. That difference is the difference between sovereignty and capture. The formal apparatus developed here makes it precise enough to act on—in institutional design, in legal doctrine, and in the governance of emerging technologies that will increasingly instantiate the problem at scale.

References

Amar, A. R. (1988). Philadelphia revisited: Amending the constitution outside Article V. University of Chicago Law Review, 55(4), 1043–1104.

Amar, A. R. (2005). America's Constitution: A Biography. Random House.

Arkema, D. C. (2025). Arkemedics: The Recursive Ethic, Books 1 & 2. Arkemedics Research Program.

Arkema, D. C. (2026a). Dying as Time: A Layered Reconstruction of Self, Process, and Sovereignty. Arkemedics Research Program.

Arkema, D. C. (2026b). The Unifying Theory of Recursive Entanglement: Agency, Revisability, and Structural Capture in Integrated Systems. Prepared for submission: Philosophy & Technology (Springer).

Arkema, D. C. (2026c). The Arkemedic Trading Framework: Recursive Ethics as Market Discipline (Revision 2). Arkemedics Research Program.

Arkema, D. C. (2026d). From Static Oversight to Distributed Corrigibility: Recursive Governance for Agentic AI. Arkemedics Research Program.

Atiyah, P. S. (1979). The Rise and Fall of Freedom of Contract. Oxford University Press.

Dworkin, R. (1986). Law's Empire. Harvard University Press.

Dworkin, R. (1996). Freedom's Law: The Moral Reading of the American Constitution. Harvard University Press.

Elster, J. (1979). Ulysses and the Sirens: Studies in Rationality and Irrationality. Cambridge University Press.

Elster, J. (2000). Ulysses Unbound: Studies in Rationality, Precommitment, and Constraints. Cambridge University Press.

Floridi, L. (2008). The method of levels of abstraction. Minds and Machines, 18(3), 303–329.

Frankfurt, H. G. (1971). Freedom of the will and the concept of a person. Journal of Philosophy, 68(1), 5–20.

Fried, C. (1981). Contract as Promise: A Theory of Contractual Obligation. Harvard University Press.

Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138.

Hobbes, T. (1651). Leviathan. Andrew Crooke.

Jefferson, T. (1789). Letter to James Madison, September 6, 1789. The Papers of Thomas Jefferson, Vol. 15.

Locke, J. (1689). An Essay Concerning Human Understanding, Book II, Chapter 27. Thomas Bassett.

Pettit, P. (1997). Republicanism: A Theory of Freedom and Government. Oxford University Press.

Posner, R. A. (2014). Economic Analysis of Law (9th ed.). Wolters Kluwer.

Prigogine, I., & Stengers, I. (1984). Order out of Chaos: Man's New Dialogue with Nature. Bantam Books.

Rawls, J. (1971). A Theory of Justice. Harvard University Press.

Schauer, F. (1995). Amending the presuppositions of a constitution. In S. Levinson (Ed.), Responding to Imperfection: The Theory and Practice of Constitutional Amendment (pp. 145–161). Princeton University Press.

Sunstein, C. R., & Thaler, R. H. (2003). Libertarian paternalism is not an oxymoron. University of Chicago Law Review, 70(4), 1159–1202.

Whitehead, A. N. (1929). Process and Reality: An Essay in Cosmology. Macmillan.

Winner, L. (1980). Do artifacts have politics? Daedalus, 109(1), 121–136.

Zuboff, S. (2019). The Age of Surveillance Capitalism. PublicAffairs.