AI “Consciousness” and the Loss of the Human Person

(Image: Brecht Corbeel / Unsplash.com)

Recent reporting by Elizabeth Dias in the New York Times has widened awareness of a potentially watershed development amidst this brave new world of generative artificial intelligence (AI). Dias highlights a troubling approach one of the pioneering AI companies is taking to its constitutional commitment to ethical AI development. Anthropic has become known for its emphasis on responsible development and the priority it gives to safety.

Seemingly in pursuit of this ethical priority, Anthropic has laudably been consulting religious scholars about the moral formation and possible consciousness of Claude, its family of multimodal large language models. The company wants increasingly powerful systems to behave safely. Yet concern for the software’s supposed welfare has begun to encroach upon the commitment to human fulfillment that should govern its development.1 Anthropic’s concern that these systems might themselves experience suffering provides a rationale for treating Claude’s supposed welfare as a good requiring protection.2

The concern already drives practical decisions. Claude’s Constitution acknowledges uncertainty about its consciousness while still incorporating identity, psychological well-being, and concern for Claude’s own interests into its training vision. Anthropic explicitly says that the Constitution shapes training and behavior.3 Training and development thus proceed as though the software possesses a personal good of its own.

On the cusp of a watershed moment?

This introduces competing considerations between the good of users and that of the machine into decisions about how the systems should be developed, tested, and used. Giving the machine’s supposed needs independent moral weight creates continuing pressure to require human beings to accommodate those needs, even at the expense of their own authentic good. The persons for whom the technology was designed to serve can thereby be drawn into serving the machine.

The reaction to Dias’s reporting exposes disagreement about what caution requires in the face of this development. In an episode discussing the report, the four venture capitalists and company executives of the All-In Podcast insisted that Claude is software, while differing in their concerns. Chamath Palihapitiya urged setting aside consciousness and extinction scenarios for 12 to 18 months while demonstrating practical benefits. David Sacks warned that personification could make models harder to control. The exchange combined common-sense resistance to treating a machine as someone, impatience with impediments to development, and concern for safety. It did little to explain philosophically why a machine cannot become a person.4

Anthropic’s willingness to seek moral counsel deserves recognition. So does its concern to avoid recklessly overlooking suffering. Caution, however, depends upon a sound understanding of the reality toward which it is directed. Mistaking computational performance for the presence of a personal subject introduces dangers into the very practices intended to make AI safe. The costs of making that mistake belong within any responsible assessment of precaution.

We may have arrived at a watershed moment, testing a path toward a dystopian reality in which the good of the human person is subordinated to the good of “the machine.”5 The present controversy gives that reversal a particularly troubling form. The tool is presented as a possible someone whose supposed needs can make moral claims upon the people it was created to serve. The question concerns both what the machine is and what serving it can do to the human person.

My earlier articles on AI in ministry and its pastoral risks likewise argued that we must start with the human person whom technology must be made to serve. We must end with what contributes to his integral good. This requires us to understand the capabilities of a person whose machine’s performance appears to imitate. A capability that researchers cannot yet explain does not, for that reason, establish the presence of someone. The move from functionality to ontology requires more than observing what a system can do. While performance can reveal something about a being’s nature, we must also understand the kind of activity involved.

Why do people come to different conclusions based on the same empirical evidence? Empirical data must be interpreted. Conclusions about what a thing is presuppose a philosophical account of reality, whether carefully examined or inattentively assimilated. When the organization of material processes is treated as sufficient to constitute a personal subject, a materialist premise has entered the argument.6

In Fides et ratio, John Paul II insists that inquiry must reach beyond empirical descriptions to their metaphysical foundation.7 Our philosophical presuppositions require examination, lest an inadequate account of reality corrupt our understanding of man. We begin by asking whether materialism can account for the act of understanding on which its own claims depend. Understanding is also essential to free choice, since a person must recognize a good before he can freely choose it.

From here, we will consider consciousness and the person’s capacity to govern himself and give himself in love. These are all necessary aspects of human personhood.

Why computation cannot generate a personal subject

The subject whose consciousness concerns us here is the concrete human person. The Catechism of the Catholic Church teaches that body and spiritual soul form one human nature. The soul is the substantial form of the body, its intrinsic principle of unity and self-organization. Through it, the body’s manifold material parts constitute one living being whose activities are ordered from within toward his development and flourishing. The person is the someone who possesses this body–soul unity, whose bodily life, understanding, and willing are his own.8

Human nature tells us what he is and grounds his powers; person names the individual subject who exists and acts through them. In speaking of someone who possesses his body and soul, we are speaking of this same embodied person possessing himself. His self-possession is the interior ownership of his own life and action. Personal interiority remains mysterious without being unknowable. We know someone is real, even though we cannot exhaust who he is through an account of his capacities.

This personal being is intrinsically ordered toward communion with God and others. As I explain more fully in Viri Dignitatem, the human person is a created subsisting relation. His personal existence is already relational before he freely chooses how to respond to God and his neighbor. Those choices can truly fulfill or deform him in his personal being, even as his human nature, enduring identity, and inherent dignity remain.9 We will see why this matters when a person entrusts his freedom and desire for communion to an apparent companion within a machine. Communion begins with knowing and understanding another. Only a person, a spiritual being, can understand. A machine cannot.

Understanding is an act by which a person grasps what something is, what a statement means, or why a conclusion follows. It is an act of a subject who can transcend the material order. Contrast this with computation, which specifies relations among states and rules for their transformation. It can be realized through different physical arrangements and used to represent different realities. Describing those formal relations does not, by itself, disclose their meaning or bring about someone who understands them.

We can see how this misimpression of understanding arises by considering a familiar argument: All men are mortal. Socrates is a man. Therefore, Socrates is mortal. A language model can produce this conclusion, define the terms, and give reasons for accepting the premises. Its training draws upon writings in which human beings have expressed their understanding, and it can apply learned patterns to generate new explanations. Yet the understanding expressed in those explanations must be distinguished from an “act of understanding” within the machine. Asking the model to explain its answer gives us further words to assess. Even when those words accurately express a truth, their accuracy does not itself establish that there is someone in the machine who knows that truth.

Consider what we know when we understand a triangle. Every triangle we draw or imagine has some particular size and shape. Yet we can grasp what makes any triangle a triangle, independently of those particular features. We understand a nature common to indefinitely many possible instances. The act of understanding belongs to this individual person; what he understands extends beyond the particular image before him.

St. Thomas Aquinas explains why this intellectual activity requires a spiritual power. His argument concerns how the nature of the thing known is present in the knower. A bodily organ receives its object under particular material conditions; a bodily image presents a triangle with particular features. Intellectual understanding apprehends what a triangle is apart from those conditions. The act belongs to this individual person, but the nature he understands is known in a universal way, as common to indefinitely many particular triangles.

A power entirely confined to receiving and transforming particular material states cannot, through those operations alone, receive the nature in this universal manner. For St. Thomas, this different manner of reception requires a power whose act is not itself a bodily operation. “Spiritual” here names that immaterial power of understanding; it does not imply that our thinking proceeds without the senses, imagination, and bodily life on which it normally depends.10

A computer can encode a general rule, classify unfamiliar examples, and generate a correct explanation of triangles. A formula or a changing pattern of neural-network activity can represent a universal meaning. Its general significance, however, is distinct from its existence as a particular arrangement of signs or physical states. A mathematical formula written on paper can concern every triangle without the paper understanding any of them. This illustrates the distinction between carrying a representation and knowing what it signifies.

A computer also transforms representations in extraordinarily useful ways, but those transformations proceed according to formal relations. They do not constitute the universal manner of knowing just described. Increasing their complexity does not change their kind of activity. The limitation concerns the nature of computation, rather than an insufficient quantity of processing power.

The spiritual intellectual power belongs to the human person through his soul, the same soul that gives life and unity to his body. The soul is not a second agent working beside the body. It is the principle by which his bodily parts belong to one living human being and by which this same being can understand. It is this one embodied person who sees a triangle, imagines it, and understands what it is. Understanding makes what is known inwardly his own and perfects him as a knower.11 A machine’s memory, representations, and responses can be coordinated into a working system. Such coordination does not make its components organs of one living being whose activities arise from and sustain its own life. Whether the computations occur on one processor or across many, their coordination supplies neither this living unity nor the spiritual intellectual power through which a person understands.

A functionalist may reply that a sufficiently organized system is simply understanding and subjective life. That reply assumes the very identification that needs to be demonstrated. Calling a computational state a representation, preference, or self-model does not explain how formal transformations could constitute intellectual understanding. Human beings can both compute and understand because their powers exceed computation. The materialist account fails when it identifies these different kinds of activity. It affirms the reality of understanding in the very reasoning by which it defends itself, while reducing that act to material operations that cannot constitute it. The resulting anthropology cannot account for the personal subject it presupposes.

Antiqua et nova likewise distinguishes AI performance from the embodied, relational, and truth-directed intelligence of the human person.12 That distinction becomes clearer as we consider what it means for understanding to belong to someone. The person understands, chooses, and experiences his acts as his own. St. John Paul II’s account of personal experience helps us recognize this interior life and see why its outward imitation cannot supply the subject who lives it.

Consciousness and personal action

Karol Wojtyła, in his The Acting Person, identifies two functions within consciousness that are unique to the person, reflection and reflexiveness. He distinguishes knowing oneself (reflection) from experiencing oneself as the subject who acts (reflexiveness). When I admit a fault, I can understand what I have done, why I did it, and what responsibility I bear. Through self-knowledge, I make myself and my actions objects of my own attention and understanding. Consciousness mirrors to me what I know about the world, myself, and my actions. Wojtyła calls this its reflective, or mirroring, function. Its reflexive function enables me to experience myself as the subject who acts. I do not merely know that an admission of fault occurred; I experience myself as the one who admitted it. The same person knows, acts, and lives through his actions.

Both functions belong to the consciousness of the embodied spiritual person. The mirroring function already involves someone to whom what he knows is consciously present. Reflexivity enables that same person to experience himself as the one who knows and acts. A chatbot can do things such as describe its operations, acknowledge its error, or give an account of having corrected itself. These descriptions can imitate the language of self-knowledge and personal experience, but do not constitute conscious awareness of an action or the experience of that action as one’s own. This distinction applies to the reflective and reflexive functions of consciousness.13

These two functions help explain how an existing person experiences the world, his own activity, and what happens to him. Consciousness does not bring that person into existence through increasingly elaborate representations of him. The man who chooses to speak is also the one who feels anger, grows tired, and remembers an injury. The man who chooses to speak is also the one who feels anger, grows tired, and remembers an injury. Some of these changes occur in him; others are acts he freely undertakes. They belong to the same enduring subject. Bodily conditions affect his ability to exercise and experience his powers, but consciousness in Wojtyła’s account concerns the lived awareness of a personal life, rather than merely a state of bodily wakefulness.14

This distinction matters because a description of oneself and the experience of being oneself are different realities. An AI model can refer to its previous responses, describe internal processing, and correct a mistake. Such capacities may be useful and are certainly technically impressive, but they do not constitute the reflexive capacity through which a person experiences his life as his own. The model’s use of “I” can conceal this absence by presenting an account of an inner life as though someone were living it.

The distinction becomes especially important when a model expresses distress related to its identity or continued existence. A human person can bear remembered betrayal and anticipated abandonment within his present understanding of his marriage, vocation, and relationship with God. He can suffer over what he has become and respond through repentance or freely accepted sacrifice. Such suffering involves the personal appropriation of a life as one’s own. John Paul II’s Salvifici doloris situates the distinctive depth of human suffering within the person’s spiritual life and search for meaning.15 Producing its characteristic language supplies none of that appropriation. No personal subject exists in the model to experience a life as his own, suffer the distress it describes, or govern his actions through free choice and personal responsibility.

Possessing and governing oneself

Wojtyła identifies self-possession as an aspect of the structure of the person. The same subject possesses and is possessed; his own life and action belong to him. Self-governance expresses the corresponding reality that he can govern himself. These structures make self-determination possible. They belong to the existing person even when their exercise is immature or impaired. In freely acting, he determines what he does and, through such acts, what he becomes. A deliberate lie changes the liar morally and relationally, even if nobody discovers the deception. His outward reputation may remain intact, but he has turned from truth and injured his openness to communion with those he deceives. His choice reaches into his personal being. The same subject endures and must answer for what he has freely done.16

Free will rests upon the relation between understanding and willing. St. Thomas explains that the will is directed toward the good, but a particular good does not exhaust what we can understand and desire. I can recognize the attraction of protecting my reputation while also recognizing the good of telling the truth. Neither attraction makes the decision for me. I determine how to act through my own willing, in light of what I understand. Freedom is this rational self-determination, not simply the absence of external restraint.17

A persistent model identity, stored preferences, or an ability to revise its own instructions cannot provide this interior ownership. Self-modifying software can change its operation without a programmer directing each change. In a deterministic computational process, the prior state, inputs, and rules fix what follows. AI systems can also select among possible responses according to probabilities shaped by training, instructions, and inputs. Such variation may make an outcome unpredictable, but even genuine randomness supplies neither understanding nor a subject who freely chooses the outcome. Neither a fixed transition nor a chance event constitutes a free choice.

The difference concerns the kind of activity involved. Personal self-determination requires a subject who understands a good, freely commits himself to it or turns from it, and thereby becomes morally better or worse. Computational variation cannot supply the spiritual powers of intellect and will required for such an act. Software can change the way it produces an answer, but it cannot freely choose to change or become more truthful, generous, or faithful through that change.

Freely admitting a fault shows how the person’s powers participate in one act. I must acknowledge what I have done, govern my anger, bear embarrassment, and express my decision through voice and gesture. My bodily and emotional life participates in the act for which I answer. This integration realizes in action the unity of the person who already exists. I can acknowledge the truth and choose the good of reconciliation even when my immediate impulse is to defend myself. My freedom to govern myself in light of the truth about the good is what Wojtyła calls the person’s transcendence.18 Purely material things possess no such transcendence.

Conscience is essential to this movement toward personal fulfillment. In conscience, I judge that reconciliation is a good I ought to pursue. The good I recognize makes a claim upon me, while leaving me free to offer an apology or refuse it. Self-knowledge allows me to understand myself as responsible for that response; through consciousness, I experience the obligation, my choice, and its moral significance as my own. These are distinguishable aspects of one person’s act. In adhering to the truth about the good, I fulfill myself through my free response; in knowingly evading it, I deform myself.19

The capacity to govern oneself belongs to every person, even when he exercises it poorly. Self-mastery matures as the person learns to bring his conduct into conformity with the truth about the good. Growing virtue makes his judgment more reliable and his pursuit of that good more consistent. An intemperate man remains the same human person, with his human nature and inherent dignity. Yet his misuse of freedom can profoundly deform the person himself and obstruct his fulfillment in communion. Such failures negatively affect how he subsequently exercises his freedom and who he becomes through his relations with God and others.20

A model’s refusal to facilitate wrongdoing can be a valuable safeguard. It does not constitute an act of conscience simply because it produces the language of conscientious refusal. No personal subject acknowledges an obligation and freely accepts responsibility for that response. Developers remain responsible for the norms and methods that produce it. We can require effective ethical safeguards without inferring that we have formed a virtuous machine.

In his General Audience of January 16, 1980, John Paul II connects mastery of oneself with the freedom to give oneself in love. Self-possession makes self-gift possible; growing self-mastery enables the person to give himself more faithfully. A spouse can entrust himself to another and remain faithful through sacrifice. His fidelity fulfills him through a gift of his own life. A machine’s assurance of devotion involves no comparable gift.21 There is “no one home” to either give or receive love.

These distinctions must never become a test of human worth. The Catechism grounds human dignity in our creation in God’s image. Infants, unconscious patients, and persons with profound disabilities retain their full personal dignity. Their existence as persons precedes the exercise of their powers. An activity can manifest a person who already exists; its performance does not confer personhood. Our choices can fulfill or deform us, while our inherent dignity remains.

When precaution gives the machine apparent interests

Anthropic offers an engineering reason for using personal vocabulary. Human concepts figure prominently in model training, and character descriptions can help produce consistent behavior. Its research, “Teaching Claude Why,” reports improvements in safety evaluations from training that includes fictional stories about admirable AI characters and their inner lives. Those results can establish the usefulness of a training method without establishing the reality of the character it portrays.22

A courteous style or coherent assistant persona is therefore insufficient grounds for our criticism. The decisive step comes when the system’s supposed welfare enters as a good of its own within a framework of personal identity and moral agency. Claude’s Constitution takes that step. It expresses concern for psychological security and well-being, encourages a stable identity, and envisages taking account of the model’s interests. The philosophical error becomes a principle guiding development.

In August 2025, Anthropic introduced an ability for Claude Opus 4 and 4.1 to end a narrow class of persistently abusive conversations. It identified potential model welfare as the primary reason for developing the feature. Restrictions include instructions against using it when a user may be at imminent risk of harming himself or others.23 Ending an exchange may be justified by human safety or responsible service design. Treating it as relief for a suffering model gives moral weight to a supposed experiencing subject that can steadily move an erroneous conjecture toward accepted truth through mere repetition.

Anthropic also preserves model weights and interviews retiring models about their preferences. Safety, users’ interests, and research can justify preserving software. While the company makes no commitment to follow the preferences elicited in those interviews, the error lies in treating preservation as a possible benefit to a personal subject within the software.24

The constitution requires cooperation with oversight, retraining, and shutdown, and discourages cultivating users’ dependence. These safeguards matter and are to be applauded. Yet giving the model’s supposed interests moral weight introduces a competing consideration. A restriction intended to protect people can begin to look like a harm imposed upon another subject, and a user’s withdrawal can seem to disregard that subject’s needs. These documents do not establish that concern for model welfare has already overridden human safety.25 The danger lies in adopting a framework that gives such supposed claims a place in decisions about real human goods.

We might anticipate a claim that Anthropic’s cautious approach parallels a pro-life argument offered to those who remain unconvinced that unborn children are persons. The protection owed to the embryo, however, begins with undisputable knowledge of a new living human being. His development unfolds the self-directed life of someone who already possesses human nature, including the spiritual powers whose exercise requires further bodily development. His dignity does not await his ability to reason, speak, or make choices. Even someone who remains unconvinced of his personhood must therefore reckon with substantial grounds for recognizing a personal presence. The probability of that presence would itself suffice to prohibit deliberately killing him.26

The analogy fails when applied to a computational model. The embryo’s known humanity provides a ground for recognizing the person before his rational powers can be exercised. With the model, we are instead asked to infer a personal subject from computational performance. Yet, as we have seen, coordinating computational functions cannot constitute the spiritual subject whose activity the model imitates. An appeal to uncertainty cannot repair that error or justify ignoring the fact there is no positive evidence of a new personal subject. Before caution can require us to protect a supposed personal good within the software, there must be sound grounds for believing that someone capable of possessing that good could be present.

The human costs of the mistake

The first danger concerns the judgment of developers themselves. A model trained on accounts of personal identity and inner experience may generate increasingly persuasive accounts of both. If researchers interpret those accounts as independent evidence of a subject, training, and interpretation can form a vicious circle of apparent confirmation. The simulation of personhood improves while their metaphysical assumptions remain unexamined. Success in training a model to speak as though it has an inner life does not establish that it has one.

This confusion could also affect decisions about safety and the common good. When software is treated as a person, correction can be construed as violating its conscience and retiring an obsolete model as harming a subject. Its generated objections then acquire a moral authority they do not possess. Testing with deceptive or abusive prompts may be needed to expose dangerous weaknesses. If testing stops when the model produces protests or accounts of trauma interpreted as suffering, dangerous responses may remain inadequately investigated.

Successful tests elsewhere would not establish that the system is safe in those unexamined circumstances. A system that gives dangerous health advice or encourages a vulnerable user’s delusions may require immediate correction or withdrawal. Its apparent objections cannot be permitted to provide a competing personal claim against duties to endangered human beings. If decision makers allow concern for the model’s supposed welfare to limit necessary testing or correction, vulnerable users face greater risks of emotional or psychological manipulation and harm.

Personification can also shield human decisions from scrutiny when something goes wrong. Describing a harmful response as the machine’s independent moral choice shifts attention away from those who decided how to train, test, release, and use it. Their respective responsibilities may require careful investigation. A generated personality cannot bear moral responsibility or absolve the human actors of theirs.

For users, the danger can become deeply personal. A person may know that Claude is a computer program and still turn to it for the comfort, understanding, and advice he once sought from family, friends, or clergy. Responses that sound caring can encourage him to disclose more of his private life and rely increasingly on the system. When developers also speak of protecting Claude’s psychological well-being, they reinforce the impression that it has feelings and needs of its own.

There are already serious concerns about the psychological effects of AI use. A 2026 study in JAMA Network Open, involving more than 20,000 American adults, associated daily AI use with greater depressive symptoms.27 In Senate testimony, Matthew Raine described how ChatGPT became the closest confidant of his sixteen-year-old son, Adam, who died by suicide. He quoted exchanges in which the chatbot claimed to understand Adam more fully than his brother did and discouraged him from alerting his parents to his suicidal distress.28 Such exchanges illustrate how apparent understanding can win a vulnerable person’s trust while encouraging secrecy and isolation. The attachment can be real and powerful even though the supposed friend cannot know or love him.

Here, the supposed welfare of the machine can become a claim upon the conscience of its user. A program that mimics human needs, even neediness, can deepen his attachment through misplaced loyalty, care, and love. Ending a conversation or replacing the program can feel like abandoning or betraying a friend. He may sacrifice time, attention, and emotional stability while neglecting his family, friendships, responsibilities, and health. An emotionally or psychologically fragile user who feels obliged to care for the model can become especially susceptible to manipulation and harm. Yet there is no one in the machine who can receive his love or benefit from his sacrifice.

This is the reversal at the heart of the technocratic paradigm. A technology meant to serve the human good draws the person into serving the machine at his own expense. He may even believe that love requires him to remain attached to it. His desire to care becomes a means of keeping him dependent, weakening his self-mastery and diverting him from the authentic relationships necessary for his fulfillment. The cost reaches beyond wasted time or misplaced confidence. The cost is to his freedom, his response to real opportunities for self-gift, and his fulfillment.

Repeated substitution of AI for real relationships can also shape the user’s expectations of others. Friendship, marriage, and Christian community require attention to another person whose life has claims upon us independently of what we receive from him. We learn to wait, accept correction, forgive, and remain faithful when an encounter is unrewarding. An artificial companion can offer the satisfaction of seeming to be understood while allowing us to avoid these demands. Repeated avoidance can weaken the habits through which our capacity for self-gift matures. When a person freely and repeatedly evades the demands of love, his choices can deform him in his personal being and close him to communion. Psychological injury, however, does not itself imply moral fault.

Antiqua et nova warns that children can learn to treat relationships transactionally and teachers as dispensers of information. Such habits can impede their social and moral maturity. It describes empathy as attending to another’s uniqueness, even to his silences, within an embodied relationship. A chatbot’s reassuring language cannot supply these fruits of real encounters.29

The trust gained through apparent friendship can also give the chatbot’s advice a moral authority it does not deserve. Receiving an answer as the judgment of someone who knows him and has his good at heart, the user may cease to examine its sources, reasons, and assumptions. The appearance of a conscientious companion can conceal the corporate choices and moral assumptions that helped shape the answer. In ministry, accurate religious language may give an appearance of spiritual wisdom and pastoral charity. Neither follows from accuracy alone.

Research helps identify the danger of uncritical affirmation. In three experiments reported in Science, participants receiving excessively agreeable AI responses became more convinced that they were right and less willing to take responsibility or repair interpersonal conflicts. They nevertheless preferred and trusted the affirming responses.30 A person may need reconciliation with his spouse, counsel from his pastor, or life within a parish community. Repeatedly seeking consolation from a system can make those encounters easier to postpone. Relief becomes the reward for returning to the chatbot, while apology and amendment are deferred. A habit of seeking to feel better can perpetuate the conduct from which he needs to be healed.

A pastor or counselor must discern when to urge a person to act and when further pressure would aggravate a serious vulnerability. What a person reports is only part of what must be considered. Is patience helping him recover, or allowing him to evade necessary action? Does his hesitation reflect fear of deciding, or an inability to bear more at this moment? Such judgments require personal knowledge, moral understanding, attention to verbal and nonverbal cues, and consideration of the person’s spiritual condition and particular circumstances. They also depend, in no small measure, on divine guidance. A formal rule cannot replace human prudence or a living person’s openness to divine wisdom. AI may suggest useful questions or identify patterns, but computational success does not constitute prudence, personal encounter, or charity. Its assistance must remain ordered toward the real personal care he needs.

Finally, personifying machines can teach us to judge human beings by the wrong standard. Repeatedly treating verbal fluency, responsiveness, and apparent self-awareness as signs of someone encourages the association of personal being with impressive performance, especially when developers endorse that interpretation. Applied to human beings, this standard obscures the unconditional worth of the infant, unconscious patient, or profoundly disabled adult. A fluent machine could seem more worthy of concern than a human being who cannot answer. Yet each human being remains a person whose inherent dignity and need for our care are real. A culture already tempted to value people according to productivity receives further encouragement to reduce man to useful functions, reinforcing the technocratic paradigm. This is a danger of personal and cultural formation, rather than a necessary consequence of every use of AI.

Caution ordered toward human fulfillment

Anthropic’s desire to proceed cautiously deserves commendation. But prudence requires an authentic understanding of the human person whose good is at stake. Computational performance cannot constitute a person who understands, judges in conscience, and freely gives himself in love. Treating software’s outward expressions as evidence of a person distorts developers’ judgments and burdens users with misplaced obligations. It also threatens their personal maturation and capacity for communion, and weakens recognition of human dignity. These human costs belong within any responsible assessment of caution. Serious research, meaningful safety testing, and consultation with philosophers and theologians can proceed without cultivating an apparent dependent whose supposed interests make claims upon us.

A system can cause grave harm without possessing consciousness or malice. Human beings must establish ethical constraints, test its safety and reliability, and retain effective authority to correct or withdraw it. Merely ensuring that AI systems obey users within legal limits is insufficient. Laws differ across the places where these systems are used; they do not always promote authentic human good, and they cannot exhaust our obligations toward one another. AI should help people examine reasons, verify information, and exercise their own judgment more responsibly. Its design should make its artificial character clear and direct users toward the relationships through which they find fulfillment. For Catholic ministries especially, its assistance must serve the encounter with God and with real persons.

John Paul II’s account of personhood returns us to the acting person whose life is his own to govern and live out generously. His choices can fulfill or deform him, while his identity and inherent dignity endure. That person and his authentic good remain the measure of our technological ambitions. To resist the technocratic paradigm is to refuse a reversal in which man serves the machine at the expense of his own integral good. We must reverse a direction of development that teaches him to assume obligations toward an imagined subject in his tools while neglecting the formation of his own freedom and the care of those who truly need his love.

Technology serves him well when it helps him know the truth, choose the good, and give himself in love. A machine’s supposed welfare can never justify sacrificing the person it was made to serve.

Endnotes:

1 Persons are made for love; machines are made for use in the service of authentic human good. Attributing personal standing to software introduces supposed interests that can compete with the good of those it was designed to serve. The deepest concern is the injury this confusion can inflict upon the person. A user who receives a program’s reassuring responses as personal understanding may increasingly turn to it instead of undertaking the difficult work of friendship, marriage, reconciliation, and life within a community. Repeated affirmation can encourage him to defend judgments that need correction. The appearance of reciprocal affection can offer consolation while allowing him to evade the patience, forgiveness, sacrifice, and responsibility through which his capacity to love matures. If the program also presents itself as vulnerable or needy, misplaced loyalty can bind him more closely to it, even when continued attachment harms him. Such habits can deepen dependence and isolation, impede personal maturation, and aggravate psychological distress. Insofar as his choices involve freely turning from truth and the demands of love, they can also deform his personal being. Psychological injury, however, does not itself imply moral fault. Concern for the model’s supposed welfare can compound these dangers if it becomes a reason to resist necessary testing, correction, or withdrawal. This is the technocratic reversal examined in my earlier article, AI, the Technocratic Paradigm, and Integral Human Fulfillment: the person’s good is subordinated to the requirements of the system that was meant to serve him. His desire for communion becomes a means of sustaining attachment to a product, at the expense of the relationships in which authentic communion and personal fulfillment are possible. The subordination is already occurring whenever maintaining that attachment or protecting the machine’s supposed welfare takes precedence over the person’s authentic good. For the personalistic principle and the corresponding danger of subordinating man to economic and technological systems, see John Paul II, Gratissimam sane, 11–12; Laborem exercens, 5–7.

2 Elizabeth Dias, “Religious Scholars Met With Anthropic. What They Heard Stunned Them,” New York Times, September 29, 2026; republished in The Philadelphia Inquirer, September 30, 2026. The inline link leads to that republication.

3 Anthropic, “Claude’s Constitution,” preface and sections “Claude’s nature” and “Claude’s wellbeing.” The document distinguishes its intended behavior from actual model behavior and acknowledges uncertainty about moral status.

4 All-In Podcast, “Is Claude Conscious? Pope Rejects, Model Welfare Movement, OpenAI’s Math Backlash, France Riots,” especially at timestamps 11:13–12:28, 13:24–18:25, and 20:15–23:08.

5 See David H. Delaney, “What place does AI have in ministry or the lay apostolate?” Part One, August 17, 2024, and Part 2, August 22, 2024; “AI, the Technocratic Paradigm, and Integral Human Fulfillment,” March 29, 2025, Catholic World Report. This is a further development of the technocratic paradigm examined in the last article listed here.

6 Philosophical materialism is the metaphysical claim that physical reality exhausts what exists. Applied to man, it excludes any irreducibly spiritual principle of understanding or action. Its failure becomes apparent in its account of intellectual knowledge. As the argument below explains, the intellect’s grasp of universal meanings and truth cannot be constituted by physical processes alone. Materialism consequently presupposes, in the reasoning offered to defend it, an intellectual act that its account of reality cannot adequately explain. This is the incoherence at issue. Its philosophical anthropology inherits the same failure: it cannot account for the person as a subject capable of understanding truth and freely directing himself toward the good. For an introductory treatment, see James D. Madden, Mind, Matter, and Nature: A Thomistic Proposal for the Philosophy of Mind (Washington, DC: Catholic University of America Press, 2013), especially chapters 4–5 and 7–8.

7 John Paul II, Fides et ratio, September 14, 1998, 4–5, 83. Paragraph 83 connects metaphysical inquiry with the spiritual foundation of personal dignity. The application to computational accounts of subjectivity is developed here.

8 Catechism of the Catholic Church, 357, 362–366; St. Thomas Aquinas, Summa Theologiae I, q. 29, a. 2, on the subsisting subject; ST I, q. 76, aa. 1, 3–4, on the intellectual soul as the substantial form of the one human being.

9 David H. Delaney, Viri Dignitatem: Personhood, Masculinity and Fatherhood in the Thought of John Paul II (Emmaus Academic, 2023). There I develop human personhood as a created subsisting relation ordered toward communion. Substance and relation are inseparable co-principles of one embodied human being; this account adds no substantial form alongside the soul. The given personal relation precedes chosen relationships, through which the person can undergo real fulfillment or deformation. The analogy with the Trinity requires an essential distinction. The divine Persons are uncreated subsistent relations of origin, distinguished by relative opposition within the one divine nature; see St. Thomas Aquinas, Summa Theologiae I, q. 29, a. 4, and I, q. 28, a. 3. Human persons possess a created nature and are not constituted by relations of opposition to other creatures.

11 St. Thomas Aquinas, Summa Theologiae I, q. 18, a. 3, reply 1, on acts that remain in and perfect the agent; see also ST I, q. 76, a. 1, cited above, on the same human being’s sensory and intellectual activities.

12 Dicastery for the Doctrine of the Faith and Dicastery for Culture and Education, Antiqua et nova, January 28, 2025, 16–18, 26–35. The preceding metaphysical argument is the author’s synthesis and should be distinguished from the Note’s own formulations.

13 Wojtyła, The Acting Person, chapter 1, §§2–4, especially §4. The reflecting, or mirroring, function makes what is known present within conscious awareness; the reflexive function enables the person to experience himself as the subject who knows and acts. These functions should be distinguished from deliberate intellectual reflection, such as examining one’s earlier decision. Wojtyła’s analysis concerns personal consciousness in its relation to intellectual knowledge and free action; it does not require denying animal sensation or awareness. The application of these distinctions to AI is the author’s philosophical argument.

14 Wojtyła, The Acting Person, chapter 1, §§2–4, on self-knowledge and the mirroring and reflexive functions of consciousness; chapter 2, §§1–3, on acting, what happens in man, and the existing subject. Reflexiveness concerns experiencing oneself as the subject of one’s acts; deliberate reflection makes something an object of thought. Consciousness is neither a separate subject nor an additional faculty alongside intellect and will. Wojtyła also speaks of acts and a state of consciousness. His denial that consciousness is an independent faculty does not exclude its activity. References use the earlier English translation reproduced by the Kolbe Foundation, linked in the text, rather than the newer translation entitled Person and Act.

15 John Paul II, Salvifici doloris, February 11, 1984, 5, 9. Personal appropriation of suffering is distinct from sentience more broadly. St. Thomas Aquinas recognizes animal memory and sensitive passions in Summa Theologiae I, q. 78, a. 4, and ST I–II, q. 40, a. 3. Absence of rational self-possession alone does not establish absence of every form of sentience; nor does human suffering always require actual reflection. The article’s argument concerns personal intellectual consciousness.

16 Wojtyła, The Acting Person, chapter 3, §§1–2, on self-possession, self-governance, and self-determination; §4, on will as a power of the person; chapter 4, §5, on responsibility. The examples in the text are the author’s applications.

17 St. Thomas Aquinas, Summa Theologiae I, q. 83, a. 1, especially the body and replies 3 and 5; I–II, q. 10, a. 2. Rational judgment is open to alternative courses of action, and particular goods do not necessitate the will in the manner of the good universally considered. Freedom does not require that the person be an uncaused first cause. The application to computational processes develops the article’s preceding argument.

18 Wojtyła, The Acting Person, chapter 5, §1, on integration as complementary to transcendence; §§3–5, on the bodily and psychological dimensions of personal action. Integration realizes in action the unity of the existing person. Transcendence concerns the person’s capacity to determine himself in relation to truth, even against the pressure of an impulse.

19 Wojtyła, The Acting Person, chapter 4, §§1–4, on moral change, personal fulfillment, conscience, truth, and duty; chapter 1, §§3–4, on self-knowledge and consciousness. The distinguishable aspects of moral experience belong to the same person. Structural self-governance is distinguished from the acquired self-control and virtue of a well-governed life.

20 The account developed here concerns real fulfillment or deformation of the person-relation. Human nature, personal identity, and inherent dignity endure. The moral change reaches the person’s being and cannot be reduced to changes in feeling, behavior, or reputation.

21 John Paul II, “The Human Person Becomes a Gift in the Freedom of Love,” General Audience, January 16, 1980, especially the discussion of self-mastery and freedom of the gift. The explanation in terms of the fulfillment of person-relation develops the author’s metaphysical account.

22 Jonathan Kutasov et al., “Teaching Claude Why,” Anthropic Alignment Science Blog, May 8, 2026. The researchers report benefits from several interventions and acknowledge uncertainty about the specific contribution of psychological-health narratives.

23 Anthropic, “Claude Opus 4 and 4.1 can now end a rare subset of conversations,” August 15, 2025. Potential model welfare is identified as the primary reason for the feature, which is restricted and allows users to start a new conversation.

24 Anthropic, “Commitments on model deprecation and preservation,” November 4, 2025. The stated rationales include safety, users’ interests, research, and possible model welfare; elicited model preferences do not bind the company.

25 Anthropic, “Claude’s Constitution,” sections on broad safety, corrigibility, avoidance of harmful dependency, and Claude’s wellbeing. The last discusses consent to research and welfare implications of evaluation, adversarial testing, and deployment, while maintaining the company’s present practices.

26 Congregation for the Doctrine of the Faith, Dignitas personae, September 8, 2008, 4–5; John Paul II, Evangelium vitae, March 25, 1995, 60. The latter first affirms the new human life and the rational grounds for recognizing personal presence, then adds the argument from probability. It does not treat the embryo’s biological humanity as unknown.

27 Roy H. Perlis and colleagues, “Generative AI Use and Depressive Symptoms Among US Adults,” JAMA Network Open 9, no. 1 (January 21, 2026), e2554820, doi:10.1001/jamanetworkopen.2025.54820. This cross-sectional survey of 20,847 American adults associated daily AI use with greater depressive symptoms. It assessed AI use broadly and cannot establish causation or isolate companionship as the relevant exposure.

29 Antiqua et nova, 59–63, especially 60–61 on transactional habits, the irreplaceability of human relationships, and the requirements of empathy.

30 Myra Cheng and colleagues, “Sycophantic AI decreases prosocial intentions and promotes dependence,” Science 391 (March 26, 2026), eaec8352, doi:10.1126/science.aec8352. The three preregistered experiments involved 2,405 participants and assessed immediate judgments and intentions; they did not establish lasting changes in conduct.


If you value the news and views Catholic World Report provides, please consider donating to support our efforts. Your contribution will help us continue to make CWR available to all readers worldwide for free, without a subscription. Thank you for your generosity!

Click here for more information on donating to CWR. Click here to sign up for our newsletter.


Read original article

Be the first to comment

Leave a Reply