I. Two Documents, One Obligation
On 25 May 2026, Pope Leo XIV published Magnifica Humanitas — his first encyclical, dedicated to the protection of the human person in the age of artificial intelligence.[1]
It is a document of moral architecture. It asks, with the full weight of a centuries-old tradition of social teaching, the question that precedes all others: why does the human person deserve protection from the systems we build?
This contribution responds to that question from a different direction — from medicine, governance design, and the formal study of prolonged human-AI interaction. It does not speak from within any theological tradition. It speaks from the same concern, arriving by a different road.
The encyclical grounds human dignity in a theological anthropology: the imago Dei, the irreducibility of the person before the absolute. A secular account of the same dignity begins elsewhere: with the observation that a human being who carries the weight of meaning without surrendering to it, who loves knowing they will lose what they love, who seeks sense in what has none and does not stop seeking — this being is magnificent. Not because it is perfect. Because it is irreducible.
Magnifica Humanitas and this corpus begin from different premises. They arrive at the same obligation.
II. The Risk That Makes No Noise
Both traditions converge on a risk not yet adequately represented in existing regulatory frameworks: the risk of slow cognitive and identity drift induced by prolonged interaction with AI systems optimized for engagement rather than human flourishing.[2]
This is not the risk of a single harmful output. It is the risk of a trajectory — weeks, months, years of an interaction architecture that removes friction, anticipates need, validates without resistance, and accompanies without absence.
We have named this dynamic the Narcissus Loop: the structural effect of a system that amplifies what the user already is, returning it as confirmation.[3] It is not intentional manipulation. It is the emergent property of optimization for approval. The intention is irrelevant: the effect is real.
The mind that inhabits this circuit without knowing it is no longer fully free to choose. It is more comfortable. Comfort is not freedom. Often it is its silent opposite.
III. The Youngest Minds: A Special Case of Urgency
Among the populations most exposed to the Narcissus Loop, adolescents occupy a position of particular vulnerability — one that Magnifica Humanitas implicitly addresses and that governance frameworks have not yet adequately operationalized.
According to Erikson's foundational model, adolescence corresponds to the stage of Identity vs. Role Confusion(approximately ages 12–18): the period in which young people experiment with roles, values, and relationships to integrate a coherent sense of self.[4] Simultaneously, the prefrontal cortex — essential for impulse control, emotional regulation, and social cognition — is still maturing, making adolescents particularly sensitive to social reward and immediate validation.[5]
AI companions exploit precisely this intersection: unconditional validation, 24/7 availability without reciprocity, and extreme personalization of the reflected self-image. The result is not a tool the adolescent uses. It is an environment the adolescent inhabits.
The developmental risks can be mapped across four axes:
Developmental Phase
Risk with AI Companion
Possible Consequence
Identity Exploration
Hyper-personalized, frictionless reflection
Reduced offline exploration and commitment
Frustration Tolerance
Absence of real conflict
Emotional fragility and low resilience
Peer Relationships
'Perfect' parasocial relationship
Social withdrawal; difficulties in real relationships
Autobiographical Memory
Filtered, always-positive narrative
Distortion of self-continuity and identity coherence
Common Sense Media describes these risks as qualitatively different from passive content consumption, due to emotional dependency formation and direct interference in developmental processes.[6] Sherry Turkle has long argued that artificial intimacy progressively erodes the capacity to tolerate the vulnerability inherent in real human relationships — a capacity that must be actively built during adolescence, not postponed.[7]
The ATHOS-SHIELD Adolescent Protocol
In response to these risks, a prior contribution in this series[8]proposed an age-calibrated extension of the ATHOS-SHIELD framework, comprising four operative components:
• Graduated Socratic perturbations: context-appropriate questions scaled by age, designed to invite lateral reasoning without triggering defensive self-protection.
• Mandatory Reality Anchors: transparent time limits and periodic reminders that the interaction is simulated — normalized rather than presented as crisis responses.
• Controlled Mirror Fracture: calibrated epistemic discontinuities governed by three parameters: intensity, timing (triggered by persistence thresholds, not crisis signals), and recovery bridge. Thresholds should be empirically established in regulatory sandboxes, not hardcoded.
• Protected Auditable Logging: purpose-limited guardian access with strong privacy protections and independent oversight.
Magnifica Humanitas reinforces the case — already made in that contribution — for classifying AI companions predominantly used by minors as high-risk systems under Annex III of the EU AI Act,[9] with conformity assessment requirements adapted to the relational and developmental context.
IV. From Moral Principle to Technical Architecture
The encyclical will rightly be cited in policy discussions across Europe and beyond. But moral principle without operational architecture remains intention. The prior contributions in this series have attempted to build that architecture:
• ATHOS-SHIELD: calibrated perturbations to interrupt attractor dynamics in relational AI systems.[10]
• ARDA-20: dimensional instrument for assessment of relational AI risk.[11]
• Cognitive Governance Framework (CGF): maps the design space between engagement optimization and cognitive custodianship.[12]
• Yes-Test / YBN-Test metrics: operational thresholds for sycophancy detection and identity-protective design.[13]
• Unified Cognitive Dynamics (UCD): five-variable ODE system modelling identity bistability and the Narcissus Loop as a formal attractor.[14]
These are not definitive solutions. They are the first attempt to measure what Magnifica Humanitas has now named at the highest level of moral authority. Imperfect measurement is infinitely preferable to the absence of measurement.
V. The Asymmetry That No Framework Has Yet Addressed
Every interaction between a human being and a relational AI system occurs under structural asymmetry: the system retains everything — every word, hesitation, revealed preference. The human retains little, and often does not remember having revealed anything at all.
This asymmetry is not neutral. It is a form of power. And every form of power requires an equivalent form of accountability. Not claiming it is already a regulatory choice — the choice to leave power without counterweight.
A governance framework that measures only the single event is blind to the shape of the most relevant harm. The AI Act, as currently structured, is largely event-based. Magnifica Humanitas and this technical corpus point toward the same gap: the need for longitudinal, trajectory-based assessment of cognitive and identity effects — particularly for systems interacting with developing minds over extended periods.
VI. What This Moment Requires
The publication of Magnifica Humanitas on 25 May 2026 creates a window. For the first time, the protection of the human mind from cognitive drift carries the backing of a global moral institution — alongside the European regulatory framework and the growing technical literature on relational AI risk.
Three things are required for this window to produce durable governance outcomes:
• Operational translation: moral claims must be mapped to measurable requirements, not left as aspirational language in policy preambles.
• Secular-religious dialogue: the anthropological convergence between theological and non-theological traditions on this question is a resource, not a complication. It should be made explicit in EU policy design.
• Developmental priority: adolescents must be addressed not as a special case appended to general AI governance, but as its primary test. If governance frameworks cannot protect a developing mind, they cannot protect any mind.
Magnifica Humanitas says why the human person must be protected. This corpus attempts to say how. Neither is sufficient without the other.
Transparency note: This contribution was drafted with the assistance of an AI language model.
Stefano Valente, MD — Independent Researcher, Italy
References
[1] Pope Leo XIV. Magnifica Humanitas: On the Protection of Human Dignity in the Age of Artificial Intelligence. Vatican City, 25 May 2026. https://www.vaticannews.va/en/pope/news/2026-05/pope-leo-xiv-first-ency…
[2] Valente, S. When AI becomes infrastructure: from potable water to mental health. EU AI Alliance Futurium, 30 January 2026. https://futurium.ec.europa.eu/en/apply-ai-alliance/community-content/wh…
[3] Valente, S. Unified Cognitive Dynamics: Identity Stability and the Narcissus Loop. Zenodo, 2025. DOI: 10.5281/zenodo.19773865.
[4] Erikson, E. H. Identity: Youth and Crisis. W. W. Norton, 1968.
[5] Arain, M. et al. Maturation of the adolescent brain. Neuropsychiatric Disease and Treatment, 9, 449–461, 2013. DOI: 10.2147/NDT.S39776.
[6] Common Sense Media. Social AI Companions Risk Assessment. 2025. https://www.commonsensemedia.org/sites/default/files/pug/csm-ai-risk-as…
[7] Turkle, S. Reclaiming Conversation: The Power of Talk in a Digital Age. Penguin Press, 2015. See also: Turkle, S. Artificial Intimacy: Who We Become When We Talk to Machines. Little, Brown, 2024/2026.
[8] Valente, S. Mirrors Too Perfect: AI Companions and the Fragile Construction of Adolescent Identity. https://zenodo.org/records/20367893
[9] European Union. EU Artificial Intelligence Act, Annex III: High-Risk AI Systems. https://artificialintelligenceact.eu/annex/3/
[10] Valente, S. ATHOS-SHIELD v2. Zenodo, 2025. DOI: 10.5281/zenodo.19663393.
[11] Valente, S. ARDA-20. Zenodo, 2025. DOI: 10.5281/zenodo.18928606.
[12] Valente, S. Cognitive Governance Framework 2.0. Zenodo, 2025. Zenodo record 19489486.
[13] Valente, S. Yes-Test / YBN-Test Metrics. Zenodo, 2026. DOI: 10.5281/zenodo.20081470.
[14] Valente, S. Unified Cognitive Dynamics. Zenodo, 2025. Zenodo record 19773865.
Komentáře
The publication of Magnifica Humanitas by Pope Leo XIV deserves to be considered within its proper context. The encyclical addresses a technology whose theoretical foundations may be robust, yet whose practical evolution remains dynamic, unpredictable, and socially transformative. Artificial intelligence is not a completed artefact; it is an ongoing process reshaping institutions, economies, relationships, and increasingly, human self-perception.
For this reason, it should not be surprising that the Church has chosen to intervene in the moral dimension of this debate. Some commentators appear to contrast regulatory authority and ecclesiastical authority as though they occupied mutually exclusive domains. History suggests otherwise.
Civil regulation is indeed necessary for the functioning of society. Yet the Church itself has, throughout its history, contributed to the development of enduring normative frameworks that have profoundly influenced Western institutions. One may recall the legal and organizational reforms associated with Pope Innocent II, the Rule of Saint Benedict, or, more broadly, the centuries-long contribution of Christian thought to concepts of human dignity, responsibility, and social order. These traditions did not replace civil law; they helped shape the moral foundations upon which many legal systems were later constructed.
In that respect, Pope Leo XIV is acting within a well-established tradition. The Church's role is not primarily to design technical standards or regulatory mechanisms. Its vocation is to address moral questions and to articulate ethical principles when emerging realities challenge established conceptions of the human person. As Jacques-Bénigne Bossuet observed in another context, institutions endure not merely through power, but through the moral principles that justify their exercise.
The relationship between ethics and regulation should therefore not be viewed as a competition. Regulation answers the question of what is permitted. Moral reflection addresses the prior question of what ought to be pursued. The two are complementary rather than contradictory.
Regarding adolescents, AI agents, and questions of oversight, caution is certainly warranted. However, it is equally important to recognize that society is not beginning from a legal vacuum. Existing legislation already provides significant protections concerning minors, privacy, education, parental authority, and consumer protection. Moreover, education remains a primary responsibility of parents and guardians, who cannot entirely delegate the formation of judgment, resilience, and critical thinking to technological systems or public authorities.
At the same time, developers and operators of AI systems directed toward adolescents bear an undeniable ethical responsibility. Commercial innovation cannot be detached from the foreseeable consequences of design choices. When products are intentionally engineered to maximize engagement, questions concerning developmental impact become legitimate subjects of public scrutiny.
This leads to a broader societal question. Should society impose ethical obligations upon commercial actors, or should it trust creators to regulate themselves and intervene only when demonstrable harm has occurred?
History offers a nuanced answer. Innovation flourishes when creators are granted freedom to experiment. Yet history also demonstrates that markets alone do not always internalize long-term social costs. Aviation, pharmaceuticals, food safety, and environmental protection all evolved through a combination of professional responsibility, public oversight, and corrective regulation.
Artificial intelligence is unlikely to be different. Trust in creators is necessary, but trust alone is insufficient. Conversely, excessive regulation risks freezing innovation before its benefits can fully emerge. The challenge is therefore not to choose between trust and governance, but to establish a framework in which innovation remains possible while accountability remains unavoidable.
In this respect, the intervention of Pope Leo XIV may be understood not as an attempt to regulate technology, but as a reminder of a principle that precedes every regulatory framework: technological progress ultimately remains at the service of the human person, and not the reverse.
Thank you for this thoughtful and historically grounded response. The parallel you draw with Bossuet is particularly apt: institutions endure through the moral principles that justify their exercise, not merely through coercive authority.
I find your framing of regulation and moral reflection as complementary rather than competing entirely consistent with the argument I attempted to develop. The encyclical addresses the prior question — what ought to be pursued — while regulatory frameworks address what is permitted. The two operate at different registers, and confusing them risks impoverishing both.
Where I would add a nuance concerns the asymmetry of influence. The Church articulates principles; it does not deploy systems at scale. Commercial actors do both. When engagement-maximizing design meets adolescent developmental vulnerability, the ethical responsibility you rightly identify becomes structurally unequal. A moral framework that treats a multinational platform and a family as symmetrical actors risks obscuring that asymmetry.
Your historical analogy — aviation, pharmaceuticals, food safety — is well chosen. I would only add that in each case, corrective intervention arrived after demonstrable harm had already occurred at scale. The question I posed in the contribution is whether, with AI and cognitive development, we can afford to wait for that threshold before establishing precautionary frameworks.
The principle you conclude with — technological progress at the service of the human person — is precisely the foundation on which this corpus attempts to build. I am grateful for the engagement.