2. The Conflict of Moral Norms

According to Kant, the supreme principle of morality is the Categorical Imperative, which he articulates through three formulations: the Formula of Universal Law (Kant, Reference Kant and Gregor1996a, 4:421), the Formula of Humanity (Kant, Reference Kant and Gregor1996a, 4:429), and the Formula of Autonomy (Kant, Reference Kant and Gregor1996a, 4:431). The third formula is not directly relevant to the aim of this article. By contrast, the first two formulas bear directly on its topic and therefore merit brief explanation.

First, no practical norm qualifies as a moral norm unless it applies universally to all rational beings. Accordingly, we may say that a moral norm is formally correct if and only if it has the form of a universal law. Equivalently, for a maxim to qualify as a moral norm, it must be capable of being adopted as a universal norm without contradiction. Hence, the Formula of Universal Law can be understood as providing the formal correctness condition for moral norms.

Second, according to Kant, every rational being is an end in itself and so must be treated as such. Thus, we may say that a moral norm is materially correct if and only if it conforms to the Formula of Humanity. Accordingly, the Formula of Humanity can be understood as providing the material correctness condition for moral norms.

What does it mean to treat a rational being as an end in itself? John Rawls (Reference Rawls and Herman2000, pp. 190–195) distinguishes between two interpretations of the Formula of Humanity: a negative one and a positive one. Each rational being holds the equal status of an end in itself, thereby establishing limits that must not be transgressed. To treat a rational being as an end in itself in the negative sense is to refrain from any action that would violate these limits. This interpretation grounds duties of right, which impose stringent, non-discretionary obligations on us — for example, the duty not to kill another person. By contrast, to treat a rational being as an end in itself in the positive sense is to actively promote two obligatory ends: one’s own perfection, and the happiness of others. This interpretation gives rise to duties of virtue, such as promoting the happiness of others. On Kant’s view, although duties of virtue, no less than duties of right, are categorical imperatives, there is an important difference between them. Duties of right are narrow (perfect) duties that admit of no latitude in their fulfilment, whereas duties of virtue are wide (imperfect) duties that allow latitude, since agents must adopt certain obligatory ends but retain discretion concerning when, how, and to what extent these ends are to be pursued.

How, then, is the Categorical Imperative related to substantive moral norms (or categorical imperatives), such as “one ought not to kill another person” or “one ought not to lie”? As noted earlier, the Categorical Imperative is the supreme principle of morality and provides both the formal and the material correctness conditions for moral norms. A substantive practical norm therefore qualifies as a moral norm only insofar as it satisfies both conditions.

Now, with the above framework of Kantian ethics in mind, let us consider the following four claims:

1. (1) Not only duties of right, but also duties of virtue are categorical imperatives.

2. (2) A duty of virtue may be fulfilled only insofar as doing so does not violate any duty of right.

3. (3) There can be circumstances in which a duty of virtue admits of no latitude.

4. (4) There can be circumstances in which a duty of virtue that admits of no latitude can be fulfilled only if the agent violates a duty of right.

To begin with, claim (1) is unproblematic. Kant is explicit that both duties of right and duties of virtue are categorical imperatives. The difference between them does not concern their normative source or authority, but rather the manner of their obligation and enforcement. With regard to duties of right, no latitude is permitted, and violations of such duties may rightly be subject to negative sanctions, such as moral blame. By contrast, duties of virtue permit latitude — at least in ordinary cases — and therefore a failure to fulfill a duty of virtue does not make an agent liable to such sanctions.

In addition, claim (2) finds strong support in Kant’s texts. Duties of right are narrow duties that admit of no latitude, whereas duties of virtue are wide duties that permit discretion in their fulfilment (Kant, Reference Kant and Gregor1996c, 6:390). The latitude characteristic of duties of virtue concerns when, how, and to what extent one fulfills them; yet this latitude must never be exercised in ways that conflict with the unconditional requirements of narrow duties. Accordingly, any such latitude must always be exercised in conformity with duties of right.

Moreover, although duties of virtue ordinarily admit of latitude, there are exceptional cases in which this latitude is severely constrained. Paradigmatic cases of urgent aid illustrate the point: When an agent is uniquely positioned to assist someone in dire need at little cost, refusal to do so rightly invites moral criticism. Admittedly, Kant does not explicitly claim that duties of virtue lose their latitude in such emergencies. Yet several prominent Kantian scholars — among them Barbara Herman (Reference Herman1993, Reference Herman2022), Thomas Hill (Reference Hill2002), and Onora O’Neill (Reference O’Neill1989, Reference O’Neill1996, Reference O’Neill2013) — have defended, either explicitly or implicitly, the view that duties of virtue may, in certain circumstances, lose their characteristic latitude, especially in cases requiring urgent assistance. Moreover, apart from exegetical considerations, there is a strong moral rationale for this view. When an agent is uniquely positioned to aid another in dire need at minimal cost, failure to act plausibly amounts to a refusal to treat that person as an end in itself. The intuitive judgement that moral criticism is appropriate in such cases is exceedingly strong. For this reason, it is desirable for Kantian ethics to recognize that no latitude is permitted in such circumstances — that is, to endorse (3).

Finally, let us turn to (4), namely, that there can be circumstances in which a duty of virtue that admits of no latitude can be fulfilled only if the agent violates a duty of right. For example, consider a situation in which an agent is the only person who can help a person in urgent need without significant difficulty. In such a circumstance, no latitude is permitted, and failing to help may therefore be subject to moral blame. Yet this requirement can come into conflict with a duty of right. The Murderer-at-the-Door scenario illustrates precisely such a conflict.

Let us suppose that S is confronted by a would-be murderer who demands information about the whereabouts of S’s friend. S recognizes the malicious intention underlying the inquiry and, as Kant puts it, finds himself in a situation in which he “cannot evade an answer of ‘yes’ or ‘no’” (Kant, Reference Kant and Gregor1996d, 8:426). In these circumstances, S may reasonably judge that the only way to save his friend from imminent death is to provide false information, thereby preventing the murderer from carrying out his plan. S’s friend is in grave and immediate peril, and S alone is in a position to provide effective assistance. Accordingly, the moral requirement to aid those in dire need applies in this case. Because this is an emergency, no latitude is permitted: Failing to aid the endangered friend would constitute a violation of the duty to assist those in desperate circumstances, and moral blame or criticism may rightly be directed at such a failure. In this case, moral criticism cannot be easily dismissed on the grounds that the relevant duty belongs to the category of duties of virtue. Under these conditions, the Murderer-at-the-Door case may be interpreted as one in which the obligation to assist those in urgent need applies and thus as involving a genuine conflict between two moral norms: the prohibition against lying (“one ought not to lie”), and the obligation to aid those in desperate circumstances (“one ought to help a person in dire need, if one can”).

Now note that the four claims cannot all be true together. Claim (3) asserts that there are circumstances in which a duty of virtue admits of no latitude; in such cases, the duty must be fulfilled in a determinate way. Given (1), this requirement has categorical force and thus binds unconditionally. Claim (4) maintains that, in some such circumstances, the only way to fulfill a duty of virtue that admits of no latitude is to violate a duty of right. Accordingly, under those conditions, the agent is required to perform an action that violates a duty of right. However, (2) imposes a strict constraint: Duties of virtue may be fulfilled only insofar as their fulfilment does not violate any duty of right. This constraint rules out precisely the kind of action that (4) identifies as required. Therefore, Kantian ethics must provide a way of resolving this inconsistency.Footnote ¹

3. Moral Norms as Social Norms

As argued in the previous section, the Murderer-at-the-Door case can be construed as a case in which two moral norms come into conflict. It is important to note, however, that Kant explicitly denies the possibility of such conflicts, maintaining that “a collision of duties and obligations is inconceivable” (Kant, Reference Kant and Gregor1996c, 6:224). In this section, I argue against this denial.

First, moral norms may be understood primarily as social norms. On Kant’s view, however, morality centres on the capacity of rational agents to autonomously direct their own lives and is therefore not primarily concerned with the social regulation of individual conduct. I do not dispute this exegetical claim. Nevertheless, my aim is not to offer a novel interpretation of Kant’s view but to articulate a more defensible version of Kantian ethics. From this revisionist standpoint, there are compelling reasons to regard moral norms as social norms. To do so is to understand them as governing relations among members of a moral community. On this view, moral norms are justified as intersubjective standards regulating relations among rational agents. Accordingly, these norms are ones that agents can demand not only of themselves but also of others. Within such a framework, social sanctions play an essential role. When a member of the moral community violates a moral norm — for example, the prohibition against killing — responses such as moral blame are justified. Without such practices, moral norms would lack the institutional and interpersonal normative force required to guide and coordinate conduct in a shared world. To claim that moral norms are primarily social norms is to affirm that they apply to agents as members of a moral community, regulating their interpersonal relations. Their effective operation depends on social forms — such as sanctioning practices — that help secure their normative authority: They rest on intersubjective justification, are upheld through positive and negative social sanctions, and are assessed intersubjectively with respect to compliance and violation. One further point worth noting is that the interpretation of moral norms as social norms aligns with Kant’s idea of “a kingdom of ends” (Kant, Reference Kant and Gregor1996a, 4:433), whose relations are governed by universally valid norms. In this respect, moral norms have an essentially relational and public dimension: They regulate how agents are to stand in relation to one another under principles that all can share.

A cautionary note is necessary here: By claiming that moral norms are primarily social norms, I do not mean that their justification is reducible to existing social practices. Rather, their justification consists in the possibility of intersubjective rational endorsement among all agents. I will defend this claim in Section 5. At the same time, I do not deny Kant’s point that we ought to perform morally right actions because they are morally right, rather than merely to avoid negative sanctions such as moral blame. This Kantian insight, however, should be distinguished from the conditions required to constitute a functioning moral community. Consider a case in which one person unjustifiably harms another but faces no negative social response. The absence of sanction neither nullifies the wrongness of the action nor releases others from their obligations. Yet systematic non-compliance, if left unchecked, undermines the practice of morality itself. When agents can no longer reasonably expect others to comply, the practical authority of moral norms as shared standards is weakened: Their ability to coordinate behaviour and guide mutual expectations is diminished. Consequently, although we must act morally because actions are right in themselves, the failure to impose negative social sanctions can jeopardize the realization of a moral community in which our relations are governed by moral norms rather than by the law of the jungle, where the strong prey on the weak. In this regard, Kant’s point is instructive: A civil condition in which public right is secured can be achieved only by regulating our mutual relations through publicly lawful external coercion (Kant, Reference Kant and Gregor1996c, 6:312). Therefore, the social enforcement of moral norms ensures that they function as publicly shared standards, securing their practical authority without undermining their intrinsic moral justification.

Second, moral norms can hardly be imposed on us by identifying all possible exceptional cases in advance and codifying them as such, since it is not feasible to foresee all such cases. For a similar reason, Allen Wood writes:

Hence, it is more reasonable to apply moral norms universally — even to cases that appear exceptional — and then assess whether there is reason to exempt agents from moral responsibility. I will elaborate on this point in the next section.

Third, moral norms function as social standards that regulate relationships among members of the moral community. To perform this role, they must be sufficiently general to apply across a wide range of circumstances. Such generality is crucial if they are to be publicly shared and readily understood by all members of the moral community. A norm that applies only to narrowly specified situations is unlikely to serve as a reliable guide amid the complexity and unpredictability of moral life. By contrast, general norms provide standards capable of guiding deliberation across diverse contexts and can be applied prospectively rather than retrospectively. Moreover, norms that lack sufficient generality cannot serve as common standards in moral deliberation or discussion. General norms enhance stability and enable mutual expectations: They allow individuals not only to anticipate their own actions but also to predict the actions of others, thereby facilitating coordination, cooperation, and trust. In short, if moral norms are to function as public standards that guide action, justify judgement, and regulate moral relations, they must possess a sufficient degree of generality.

Fourth, within Kantian ethics, there can be no hierarchy among moral norms or duties. Moral norms must be categorical imperatives, unconditionally obligatory for all rational beings as such. If, when a conflict arises between two norms, N₁ and N₂, N₁ is said to take precedence over N₂, this implies that N₂ is obligatory only on the condition that it does not conflict with N₁. In that case, N₂ would not be an unconditional categorical imperative. The claim that a hierarchy exists among moral norms is therefore incompatible with Kant’s core commitment to their unconditional normative force.

A related point concerns the absence of any higher-order principle capable of resolving genuine conflicts among moral norms. Suppose that norms N₁ and N₂ come into conflict. For this to constitute a genuine conflict, both must qualify as genuine moral norms — that is, they must satisfy the formal and material correctness conditions. Whether a norm meets these conditions is ultimately determined by the supreme principle of morality. Accordingly, insofar as N₁ and N₂ both satisfy them, there can be no further principle that adjudicates conflicts between them. In this respect, it is noteworthy that both norms may be indispensable for treating rational beings as ends in themselves. For example, the norms that one ought not to lie and that one ought to help those in dire need, when one is able to do so, may each be necessary for fulfilling the requirement to respect humanity as an end in itself. Each norm therefore can be equally justified by the supreme principle of morality. It is true that, insofar as duties of virtue ordinarily admit of latitude, their fulfilment must remain within the bounds set by duties of right. Violating a duty of right does not fall within the scope of permissible latitude. In this qualified sense, duties of right may be said to take priority over duties of virtue. However, this priority holds only where latitude is in fact permitted. In exceptional cases in which duties of virtue admit no latitude, as argued in the previous section, duties of right do not take priority over such duties of virtue.

For these reasons, it is reasonable to recognize that genuine conflicts among moral norms can arise in certain exceptional cases. On this revisionary Kantian account, the Murderer-at-the-Door case is best understood as involving a genuine conflict between two moral norms: the prohibition against lying, and the obligation to aid persons in desperate circumstances when the agent alone is able to do so.Footnote ²

4. The Default-and-Exemption Account of Moral Responsibility

As previously noted, Kant insists that the injunction “one ought not to lie” is a categorical imperative that admits of no exceptions. On his view, agents are required not to lie even in exigent circumstances, such as the Murderer-at-the-Door case. This uncompromising stance, however, conflicts with the widely shared intuition that lying may be morally permissible in exceptional cases of this kind. In what follows, I argue that Kant’s fundamental claim — that “one ought not to lie” is a categorical imperative — can nonetheless be reconciled with this intuition. The key to this reconciliation lies in the default-and-exemption account of moral responsibility; this section therefore outlines that account.

As discussed in Section 2, the Formula of Universal Law provides the formal correctness condition for moral norms: A norm is formally correct if and only if it has the form of a universal law — that is, if and only if it can be applied without exception to all rational beings. Similarly, the Formula of Humanity provides the material correctness condition for moral norms: A norm is materially correct if and only if it conforms to the Formula of Humanity. Moreover, the Formula of Humanity admits both negative and positive interpretations. On the negative interpretation, we are required not to infringe upon the equal status of any rational being as an end in itself; on the positive interpretation, we are required to promote the happiness of rational beings.

What, then, is an example of a moral norm that satisfies these correctness conditions? One such example is the prohibition on killing another person. This norm satisfies both the formal and material correctness conditions. To kill another person is to violate their basic right to life, thereby infringing upon a duty of right grounded in the Formula of Humanity, construed negatively. For this reason, killing is morally impermissible, and the corresponding duty applies to all rational beings without exception. Nevertheless, there may be exceptional cases in which an agent acts contrary to such a norm without thereby being morally responsible. Consider, for instance, a case in which an agent S kills another person T in self-defence against T’s attack. In such a case, S may be exempt from moral responsibility. Along these lines, it can be argued that if S kills T, the default condition is that S is morally responsible for T’s death. However, if S acts in self-defence against T’s aggression, S may be exempt by virtue of a justifiable reason for exemption.

Exemptions from moral responsibility may also admit of degrees. Suppose S kills T in self-defence, though T’s threat is not genuinely life-threatening. In such a case, S’s response may be excessive, making full exemption difficult. While S may bear reduced responsibility for T’s death, complete exemption would be unjustified.

Furthermore, as noted in Section 3, moral norms can hardly be specified by identifying in advance all possible exceptional cases. The range of situations in which one might kill in self-defence is too diverse to be exhaustively anticipated. It is therefore more plausible to treat moral norms as universally applicable by default — even in cases that initially appear exceptional — and to determine, in light of particular circumstances, whether the default condition of responsibility is overridden.

These considerations support understanding moral responsibility in terms of the default-and-exemption account. On this account, when a rational agent violates a moral norm, the default condition is that the agent is morally responsible for the violation. This condition may, however, be overridden in exceptional cases in which the agent has a justifiable reason for exemption. In such cases, an agent who violates a moral norm may be wholly or partially exempt from responsibility. This constitutes the exemption condition.

Three merits of this account deserve particular attention.

First, it preserves Kant’s fundamental commitment to the universality of moral norms while accommodating the intuition that agents may, in exceptional circumstances, be exempt from responsibility despite having violated them. The default condition explains why moral norms have the status of categorical imperatives: Whenever an agent violates such a norm, the presumption is that the agent is responsible, and it applies universally. At the same time, the exemption condition explains how agents may be exempt from responsibility in exceptional circumstances — namely, when there are overriding reasons for exemption. In this connection, it is crucial to distinguish between the universal authority of a moral norm and the blameworthiness of its violation. An analogy with epistemic norms is instructive here. Consider modus ponens, a basic rule of inference that all rational beings ought to follow. We do not regard young children or individuals with significant cognitive impairments as epistemically blameworthy for failing to conform to it. This does not show that the rule fails to apply to them; rather, they may be exempt from epistemic blame for justifiable reasons. The same holds for moral norms. Cases in which it is inappropriate to hold an agent responsible for violating a moral norm do not entail that the norm itself is inapplicable.

Second, this account explains why agents who violate moral norms bear the burden of providing reasons for exemption. Unless agents — or their defenders — can offer justifiable reasons that override the default condition, that condition remains in force. For example, if agent A lies to agent B, A may be required to justify the action; in the absence of such justification, A remains subject to moral criticism. Exceptional cases are therefore best understood not as instances in which moral norms fail to apply, but as cases in which agents may be exempt from responsibility. If norms were genuinely inapplicable in such cases, it would be difficult to explain why violations continue to invite moral scrutiny or why agents are expected to offer justifications. The very practice of demanding such reasons presupposes that the relevant norms remain in force.

Third, this account allows that cases that appear to involve conflicts between moral norms — such as the Murderer-at-the-Door scenario — may constitute genuine moral conflicts. Even in such cases, the present account can explain which course of action is more reasonable by appealing to reasons for exemption. A more detailed account of this claim will be provided in the next section.

It is worth emphasizing that the default-and-exemption account is at least consistent with Kantian ethics and can therefore be incorporated within a Kantian framework. Kant himself frames moral responsibility in terms of imputability. “Imputation (imputatio) in the moral sense is the judgment by which someone is regarded as the author (causa libera) of an action, which is then called a deed (factum) and stands under laws” (Kant, Reference Kant and Gregor1996c, 6:227). To impute an action to an agent is thus to regard that agent as its author and to hold them responsible for it. Such a judgement implies that the action stands under laws — specifically, moral laws. Accordingly, insofar as an action accords with or violates a moral norm, the agent is subject to moral evaluation and is thereby assessable as responsible. The present account is consistent with this Kantian conception of imputability, while providing a more fully articulated framework for the attribution of responsibility. The norm “one ought not to lie” functions as an unconditional categorical imperative. When a rational agent violates such a norm, the imputation of responsibility constitutes the default condition. However, if there are justifiable reasons for exemption, the exemption condition requires that we either withdraw the imputation or assign a diminished degree of responsibility, depending on the nature of those reasons.Footnote ³

There are also independent reasons to think that a Kantian framework can accommodate degrees of responsibility. Kant distinguishes between positive and negative freedom (Kant, Reference Kant and Gregor1996a, 4:446). Positive freedom may be understood in terms of the rational capacity to act in accordance with moral norms (see Allison, Reference Allison2019, p. 455). On this view, moral norms apply to us without exception because we possess this capacity; it would be pointless to impose moral requirements on beings that lack it altogether. Negative freedom, by contrast, can be understood in terms of the absence of determination by alien causes (see Kant, Reference Kant and Gregor1996a, 4:446). Anything outside the rational will can count as such an alien cause, including pathological desires and inclinations. We morally criticize certain actions in order to prevent their recurrence. However, in the case of actions that lie beyond an agent’s rational control, demanding that the agent refrain from them is futile. For example, if a kleptomaniac steals something due to an irresistible urge, they may be regarded as lacking negative freedom with respect to this act. Negative freedom therefore requires not only the absence of illegitimate external constraints but also the absence of internal compulsion or other pathological factors. In other words, agents lack negative freedom with respect to their actions when those actions are determined by internal or external factors beyond their rational control. Along these lines, it can be argued that moral responsibility requires not only positive freedom but also negative freedom. Notably, whereas positive freedom is an all-or-nothing capacity, negative freedom admits of degrees. An agent may have had no rational control at all over the action they performed, may have lacked such control to a significant extent, or may have lacked it only partially.Footnote ⁴

The above distinction between positive and negative freedom fits well with the default-and-exemption account. If an agent possesses positive freedom, the default condition of responsibility applies. However, insofar as the agent wholly or partially lacks negative freedom with respect to a particular action, the exemption condition may be invoked. Moreover, any moral theory that fails to accommodate degrees of responsibility is implausible. It is counterintuitive to attribute the same degree of responsibility to a child or a cognitively impaired individual as to a fully competent adult. Likewise, it is implausible to treat an action performed under a threat of death as equally blameworthy as one performed entirely voluntarily.

My account may be regarded as a version of Kantian ethics for three additional reasons. First, it accepts the supreme principle of morality, the Categorical Imperative, in its three formulations — the Formula of Universal Law, the Formula of Humanity, and the Formula of Autonomy — and interprets the first two as providing the formal and material correctness conditions for moral norms. Second, it endorses the Kantian view that moral evaluation depends not on consequences but on the reasons for which an agent acts. What matters are the agent’s reasons — those that justify their actions to others. Since consequences often lie beyond our control, they are not directly relevant to moral assessment. Third, my account upholds the Kantian requirement of impartiality: Moral reasons must have the same normative force for all rational beings. If an agent has a justifiable reason for exemption in a given situation, then any agent in a relevantly similar situation must be equally exempt. Exemptions must therefore be universalizable. For these reasons, my account can be regarded as a version of Kantian ethics.Footnote ⁵

5. Categorical Imperatives and Its Permissible Violations

Drawing on the default-and-exemption account of moral responsibility articulated in the previous section, this section explains why lying may be permissible in certain exceptional circumstances, while preserving the Kantian claim that moral norms are categorical imperatives.

Consider again the Murderer-at-the-Door case. S faces only two options: either to lie in order to save his friend, or to disclose his friend’s whereabouts, thereby enabling the murderer to kill him. Both alternatives involve moral wrongdoing. The first violates the prohibition against lying; the second violates the duty to aid those in dire need. Thus, each option entails a violation of a moral norm, yet S must choose between them. The central question, then, is which option S ought to choose. In circumstances of this kind, the best S can do is to choose the morally less bad option available. Which option, then, is morally less bad?

The first option — lying — fails to accord with the Formula of Humanity. Let me explain. According to Kant, lying consists in making an untruthful statement with the intention of deceiving another into believing it to be true. He calls lying in this sense lying “in the ethical sense” (Kant, Reference Kant and Gregor1996c, 6:430). In his view, lying in the ethical sense requires that a statement be made, and that the statement be believed to be false by the person who makes it.Footnote ⁶ It is not required, however, that the statement be false. Thus, a lie in the ethical sense (henceforth, simply a lie) may turn out to be true. In addition, the moral status of a lie does not depend on whether the deception succeeds; even if no one is misled, the act still counts as a lie in the ethical sense. Moreover, some lies can infringe upon another person’s juridical rights. For instance, if one falsely claims that a contract has been legally concluded when it has not, thereby causing legal harm to another party, this constitutes a violation of that party’s juridical rights. Yet many lies bear no relation to the violation of juridical rights. Thus, the mere fact that certain lies can undermine juridical rights does not explain why all lies are morally impermissible. A deeper question, then, concerns the justification of the moral norm “one ought not to lie.” What grounds the claim that every lie is morally impermissible? This question can be addressed by appeal to Kant’s Formula of Humanity.

Rational agents must be free to set their own permissible ends, to determine the means appropriate to their realization, and to act upon those determinations, provided that these remain compatible with the equal freedom of others. Lying violates this requirement, for it constitutes an intentional attempt to distort another person’s understanding of reality, thereby undermining their capacity to make rational choices on the basis of accurate information. What is decisive here is not whether this impairment succeeds, but rather that the maxim of lying subordinates another’s rational agency to one’s own ends. To interfere with another’s autonomy in this way is to treat them as a mere means rather than as an end in itself.

Moreover, given our epistemic limitations, we are necessarily dependent on others for information — a dependence that underwrites the indispensable role of linguistic communication. The point is not merely that, if lying were generally permitted, the trustworthiness of testimony would collapse. Rather, the practice of testimony itself presupposes a norm of truthfulness, such that a maxim permitting lying cannot be coherently willed as a universal condition of shared epistemic agency. As a result, each person’s capacity to choose appropriate means to their ends on the basis of true information would be undermined. In this context, Kant writes:

Along these lines, Kant argues that truthfulness is “a human being’s duty to everyone,” and that every lie is “a wrong inflicted upon humanity generally” (Kant, Reference Kant and Gregor1996d, 8:426). To put the point another way, every lie is morally impermissible because it violates at least a “right of mankind,” or a “right of humanity,” understood as a non-juridical right grounded in the conditions of mutual rational interaction. Therefore, if S in the Murderer-at-the-Door case takes the first option, he not only infringes upon the would-be murderer’s capacity for rational decision-making but also violates a right of humanity.Footnote ⁷

Let us now turn to the second option. Disclosing the required information would foreseeably result in the death of an innocent person — a grave and irreversible harm. In addition, providing information that enables the commission of a wrongful act plausibly counts as assisting that act, and assisting serious wrongdoing is itself morally impermissible. Accordingly, one ought not to provide information required for the commission of murder. In such circumstances, the only way to avoid providing such assistance may be to lie — for example, by falsely claiming ignorance.

If these considerations are correct, S — or a defender of S’s action — may argue that lying is the morally less bad option in the Murderer-at-the-Door case. By assumption, S must choose either to disclose his friend’s whereabouts to a would-be murderer or to lie. In the former case, his friend is likely to be killed. From the standpoint of each person’s status as an end in itself, the most serious infringement one can suffer is being killed by another. In other words, killing a person is the most serious violation of their status as an end in itself, as it inflicts an irreversible and profound harm. Moreover, disclosing the requested information would amount to assisting in serious wrongdoing. In the latter case, by contrast, S’s friend is likely to be saved. Admittedly, this option involves interfering with the murderer’s rational decision-making. However, interfering with the rational deliberation of an agent who intends to commit murder, though a genuine wrong, is less severe. It is rare for a single instance of misinformation to undermine a person’s life as a whole. Consequently, deliberately providing false information that undermines a person’s capacity to make a particular rational decision on the basis of truth is, in most cases, a far less serious infringement. Moreover, exempting agents from moral responsibility for lying in highly exceptional cases would not plausibly undermine the general trustworthiness of linguistic communication. In light of these considerations, it is difficult to maintain that lying is morally worse than allowing the murder of an innocent person. Accordingly, the fact that lying is unavoidable in order to prevent complicity in serious wrongdoing constitutes a compelling reason for exemption. One additional point worth mentioning is that, as noted in the previous section, a Kantian framework can accommodate degrees of responsibility. Along these lines, S — or a defender of S’s action — may argue that lying is the morally less bad option in this case.

If this line of argument is sound, S cannot be criticized on the grounds that he ought to have chosen a better alternative. To hold S morally blameworthy, one must be able to maintain that he ought to have acted otherwise. Yet it is difficult to argue that allowing the murder would have been the morally better option. Consequently, even if S lies, he cannot be faulted for failing to make a better choice. In such a case, S possesses a justifiable reason for exemption: There is no better alternative to saving his friend’s life. This reason overrides the default condition of responsibility for lying. Accordingly, S may be exempt from moral responsibility for his lie. To say that S is exempt from responsibility for lying is, in turn, tantamount to saying that the lie is morally permissible. The default-and-exemption account of responsibility thus allows us both to preserve the Kantian view that the norm “one ought not to lie” is a categorical imperative and to acknowledge that lying may be permissible in exceptional cases such as the Murderer-at-the-Door.

It is worth emphasizing that this account is not a form of consequentialism. As noted in the previous section, it affirms the core theses of Kant’s moral philosophy. While there may be exceptional cases in which an agent is exempt from responsibility despite violating a moral norm, such norms retain their categorical normative force for all persons. The fact that determining whether the exemption condition is satisfied may involve comparing options to determine which is less morally bad does not entail a consequentialist evaluation of the relative priority of moral norms. Such comparisons do not determine which moral norm takes precedence; rather, they provide reasons for holding that, in rare circumstances of genuine moral conflict, agents may be exempt from responsibility even though they have violated a moral norm. Moreover, whether the exemption condition is satisfied in a given case does not depend on the actual consequences of the action. What matters instead is whether, at the moment of decision, it was reasonable for the agent to believe that the exemption condition obtains.

At this point, one might wonder whether my account can provide a universal, public, and objective criterion for determining the circumstances under which an exemption is appropriate. A Sellarsian coherence theory of justification offers the resources to address this concern.

The coherentist insight is that we have no rational alternative but to evaluate the truth or justification of a belief within our conceptual framework. Two points deserve emphasis. First, as Ludwig Wittgenstein (Reference Wittgenstein, Anscombe and von Wright1969, §§341–343) observes, genuine doubt is possible only against a background of beliefs that are not simultaneously doubted. Second, as Kant insists, reason is the highest court of appeal for all questions concerning itself and its principles. Accordingly, reason must answer such questions from its own resources.Footnote ⁸ Put differently, our conceptual framework provides the norms, criteria, and rules for defending or criticizing any claim, and so it is inevitable that demands for justification be addressed within that framework. Here, a “conceptual framework” may be understood as a framework within which genuine doubt is settled and demands for justification are met.

We may further adopt a Lehrerian account of coherence. Coherence is often characterized in terms of how the elements of a belief system fit together, but beyond such metaphorical descriptions, no generally accepted definition is available. In response to this difficulty, Keith Lehrer offers an illuminating proposal: Coherence can be understood in terms of “answering all objections” or “beating all competitors” (Lehrer, Reference Lehrer2000, p. 129). One’s belief that p coheres with one’s background system just in case one can answer all objections raised against it on the basis of that system. However, as Wilfrid Sellars (Reference Sellars1963) argues, the very concept of justification arises from our social practices of giving and asking for reasons. Lehrer’s account should therefore be recast in intersubjective terms. On the Sellarsian version of coherence theory that I defend, we are justified in believing that p just in case we can answer all objections raised against it within our social practice of demanding justification and responding to such demands. Although this account has primarily been developed for epistemic justification, the underlying insight applies equally to practical justification: When we are required to justify actions or practical norms, we likewise have no rational alternative but to do so within our social practice of justification.Footnote ⁹

With this social practice model of justification in place, we can see how a Kantian framework accommodates intersubjective assessments of the exemption condition of responsibility. As argued in Section 4, the default-and-exemption account is at least consistent with Kantian ethics and can therefore be incorporated into such a framework. In particular, my account upholds the Kantian requirement of impartiality: Moral reasons must have the same normative force for all rational beings. If an agent has a justifiable reason for exemption in a given situation, then any agent in a relevantly similar situation must be equally exempt. Exemptions must therefore be universalizable. Accordingly, insofar as it is successfully established within our social practice of justification that a reason offered in support of an exemption is universalizable, that reason qualifies as a justifiable reason for exemption. Therefore, in the Murderer-at-the-Door case under consideration, insofar as S — or a defender of S’s lie — successfully argues that he has a jusifiable reason that exempts him from responsibility for the lie, S is not to be held responsible for it.Footnote ¹⁰

Finally, to avoid potential confusion regarding my claim that lying may be “permissible” in certain exceptional cases, we may distinguish between two senses of permissibility: strict norm-relative permissibility, and responsibility-relative permissibility. This distinction may be clarified as follows. For Kant, permissibility is not a comparative notion grounded in the absence of better alternatives; rather, it is a categorical status determined solely by conformity with the moral law. An action that violates a moral norm remains impermissible, regardless of whether all available options are bad. Moreover, Kant distinguishes between the objective moral status of an action and the agent’s culpability. The fact that an agent is not morally blameworthy does not entail that the action itself is permissible. He allows that agents may be excused — or at least partially exculpated — on account of ignorance, coercion, or severe limitations on their capacity to act, while maintaining that the action remains contrary to duty.

I do not, however, accept Kant’s claim that genuine conflicts of duty are impossible. Any plausible moral theory must acknowledge tragic cases in which wrongdoing is unavoidable. To address such cases, I adopt the default-and-exemption account of moral responsibility. On this account, when a rational being violates a moral norm, the default condition is that they are morally responsible for the violation. This default condition can be overridden in exceptional cases where the agent has a justifiable reason for exemption. In other words, an agent who violates a moral norm for such a reason may be — at least partially — exempt from responsibility. This constitutes the exemption condition of moral responsibility. Within this framework, we may distinguish between two senses of permissibility. The first is strict norm-relative permissibility: An action is permissible only if it accords with moral norms. In such permissible cases, the agent bears no moral responsibility. This corresponds to the standard understanding of permissibility, according to which to say that an action is permissible is simply to say that it is not morally wrong. The second is responsibility-relative permissibility: An action may count as permissible under exceptional circumstances in which no better option is available and the agent is exempt from responsibility for performing it. In this sense, an action may be practically permissible — that is, one may perform it without being subject to moral blame, even though it remains normatively wrong.

If responsibility is the primary locus of moral appraisal in practice, it is natural to introduce a derivative evaluative category for actions that, while violating norms, are nonetheless exempt from blame. Our moral practice implicitly relies on this responsibility-sensitive notion of permissibility, especially in the context of practical deliberation. Consider an agent who is faced with two morally bad options: In such a case, we may appeal to the considerations that it is rational to choose the lesser evil, together with the possibility that the agent is exempt from responsibility for doing so. On this basis, we can legitimately say: “You may choose the morally less bad option.” To recommend such a choice is, in effect, to confer permissibility upon it. Moral advice under tragic conditions thus presupposes a conception of permissibility tied to responsibility and justification rather than strict norm-conformity. To recommend an action in such circumstances is, practically speaking, to permit it. Any adequate moral theory should be able to accommodate this intuitive judgement.

In short, we can distinguish two levels of moral evaluation: one assessing conformity to moral norms, and the other assessing responsibility in concrete circumstances. The default-and-exemption account allows us to maintain that moral duties remain strict and exceptionless at the level of norms, while responsibility remains sensitive to the agent’s practical situation — including cases in which every available option involves a violation of a moral norm. This distinction enables us to separate strict normative permissibility from responsibility-relative permissibility: that is, what is permissible at the level of moral norms from what is permissible at the level of practical deliberation. Crucially, allowing less bad choices does not implicitly rank moral norms or duties. The comparison operates at the level of practical decision-making under conditions of norm violation, not at the level of the norms themselves. Moral norms remain strict and exceptionless qua norms. By distinguishing between these two senses of permissibility, my proposal preserves Kant’s strict view of the categorical authority of moral norms while offering a more realistic account of responsibility in exceptional cases of moral conflict.Footnote ¹¹