Magnifica Humanitas – Pope Leo’s Take on Intelligence – Artificial and Otherwise
The first thing to understand about Pope Leo XIV’s new encyclical, Magnifica Humanitas, is that it is not really about artificial intelligence. It is about power. It is about who has it, who does not, who is watched, who is classified, who is automated, who is replaced, who is forgotten, and who gets to decide what “human” means when machines can imitate human judgment, speech, creativity and even moral reasoning.
The document, issued by the Holy See under the title Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence, frames AI not as a discrete technology-policy problem, but as the newest version of an old human temptation: the desire to build a Tower of Babel. The encyclical contrasts Babel with Nehemiah’s rebuilding of Jerusalem. Babel represents technical unity without moral community, efficiency without dignity, and power without restraint. Jerusalem represents shared responsibility, subsidiarity, pluralism and rebuilding institutions around the dignity of persons rather than the dominance of systems. The document’s central claim is that “technology is never neutral,” because it “takes on the characteristics of those who devise, finance, regulate and use it.” Leo XIV, Magnifica Humanitas ¶ 9, Holy See Press Office (May 15, 2026).
That is a profoundly cyber-relevant claim. Cybersecurity law and policy have spent decades trying to separate tools from intent. A port scanner may be lawful research or reconnaissance. Encryption may be privacy or concealment. A vulnerability disclosure may be a public service or extortion. AI collapses those distinctions further. The same model can summarize medical records, write ransomware, detect fraud, fabricate evidence, identify battlefield targets, create deepfakes, optimize hiring or silently score a person’s creditworthiness, employability, insurability or loyalty. AI is not simply software. It is a decision infrastructure.
The encyclical is therefore more demanding than most governmental AI policies. Governments are largely asking whether AI systems are safe, competitive, explainable, auditable, non-discriminatory or strategically useful. The Vatican asks a prior question: what kind of human beings and what kind of society are these systems designed to produce?
That distinction matters. The United States, the European Union, the United Kingdom, China, Canada, Japan, South Korea, the OECD and the United Nations are all doing something about AI. But they are not doing the same thing, and very few are asking the deeper question raised by Magnifica Humanitas: whether the current AI race is governance at all, or merely Babel with better compute.
The United States: Innovation First, Governance Second
The current U.S. approach is fragmented, contested and strongly tilted toward innovation, competition and national-security advantage. President Trump’s Executive Order 14179, “Removing Barriers to American Leadership in Artificial Intelligence,” revoked prior AI policies viewed as barriers to American AI leadership and directed the federal government to remove obstacles to U.S. dominance in the field. Exec. Order No. 14,179, Removing Barriers to American Leadership in Artificial Intelligence, White House (Jan. 23, 2025)/.
The administration’s America’s AI Action Plan is similarly organized around accelerating innovation, building AI infrastructure, and leading international diplomacy and security. America’s AI Action Plan, White House; AI.gov, President Trump’s AI Strategy and Action Plan.
There are important exceptions. OMB Memorandum M-25-21 continues to require federal agencies to accelerate AI adoption while maintaining safeguards for privacy, civil rights and civil liberties. Off. of Mgmt. & Budget, Memorandum M-25-21, Accelerating Federal Use of AI Through Innovation, Governance, and Public Trust (Apr. 3, 2025). NIST’s AI Risk Management Framework remains the closest thing to a national baseline for private-sector AI risk governance, but it is voluntary, not a comprehensive federal statute. Nat’l Inst. of Standards & Tech., Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (Jan. 2023).
The United States is also pursuing AI literacy and workforce readiness. Executive Order 14277 directs AI education and proficiency initiatives for American youth, and Executive Order 14278 focuses on workforce programs and skilled-trade jobs of the future. Exec. Order No. 14,277, Advancing Artificial Intelligence Education for American Youth, White House (Apr. 23, 2025); Exec. Order No. 14,278, Preparing Americans for High-Paying Skilled Trade Jobs of the Future, 90 Fed. Reg. 17,697 (Apr. 28, 2025).
But the U.S. model remains largely procedural and strategic. It asks whether government can use AI responsibly, whether industry can innovate rapidly, whether American firms can beat Chinese firms, whether agencies can procure safely, whether workers can be trained, and whether state AI laws should be preempted. It does not yet ask, at least not systematically, whether AI systems undermine human freedom by converting the person into a datapoint, a target, a risk score or a behavioral prediction.
That is precisely where Magnifica Humanitas is sharper. It warns that massive data collection and algorithmic systems create “the power to profile, predict and influence behavior,” and that decisions about credit, employment and essential services can become instruments of discrimination and control unless there are “clear rules, transparency, the possibility of recourse and proportionate limits.” Leo XIV, Magnifica Humanitas ¶ 171.
In American legal terms, the encyclical is calling for due process, contestability, purpose limitation, anti-discrimination controls, explainability and limits on surveillance capitalism. The U.S. has pieces of that architecture in sectoral laws, consumer-protection enforcement, civil-rights law, privacy law, procurement policy and voluntary risk frameworks. It does not yet have a unified national AI law.
State Law
The states are filling that void. Colorado enacted one of the most comprehensive state AI governance statutes, focused on high-risk AI systems and consumer protections, although implementation has been delayed and amended. Colo. S.B. 24-205, 74th Gen. Assemb., 2d Reg. Sess. (2024). California has moved on transparency through laws such as AB 2013, requiring public-facing generative AI developers to post training-data documentation, and SB 53, the Transparency in Frontier Artificial Intelligence Act, addressing safety disclosures and frontier-model risks. Cal. A.B. 2013, 2023–2024 Reg. Sess. (Cal. 2024).
The Vatican’s critique is that these laws, while useful, are still mostly about documentation, risk, safety, fairness and disclosure. They do not necessarily prevent the deeper problem: a society in which the person becomes machine-readable, machine-predictable and machine-governed.
The European Union: Rights-Based Regulation With Teeth
The European Union comes closest to converting the Vatican’s moral vocabulary into enforceable law. The EU AI Act entered into force on August 1, 2024, with full applicability generally scheduled for August 2, 2026, subject to earlier application for certain prohibited practices and AI literacy requirements. Regulation (EU) 2024/1689, Artificial Intelligence Act, 2024 O.J. (L 2024/1689); European Comm’n, AI Act.
The EU model classifies AI systems by risk. Some practices are prohibited. High-risk systems are subject to obligations involving risk management, data governance, transparency, human oversight, accuracy, robustness and cybersecurity. General-purpose AI models receive separate treatment. This is recognizably legal architecture rather than aspirational guidance.
In the language of Magnifica Humanitas, the EU is trying to prevent Babel by zoning the construction site. It is saying that some AI uses are incompatible with fundamental rights; some require permits, controls and inspection; some require transparency; and some are largely left alone. The encyclical would likely welcome that structure, but it would also press further. It would ask whether risk classification alone is sufficient when AI infrastructure is privately controlled, globally concentrated, extractive of labor and energy, and embedded into the architecture of speech, work, education and war.
That is the encyclical’s most important contribution. It is not satisfied with compliance. It wants conversion. It says social justice is not something to be bolted onto AI after deployment, but must shape design “from the outset.” Leo XIV, Magnifica Humanitas ¶ 109.
The United Kingdom: Pro-Innovation, Public-Sector Adoption and Regulatory Pragmatism
The United Kingdom has chosen a less centralized, more pro-innovation approach. Its AI Opportunities Action Plan, published January 13, 2025, emphasizes AI infrastructure, compute, public-sector adoption, growth, skills and national competitiveness. U.K. Dep’t for Sci., Innovation & Tech., AI Opportunities Action Plan (Jan. 13, 2025).
The U.K. model is pragmatic. Rather than creating an EU-style comprehensive AI act, it relies more heavily on existing regulators, sectoral expertise and national industrial policy. The government’s “one year on” reporting emphasizes adoption, BridgeAI, robotics hubs and efforts to accelerate deployment in priority sectors. U.K. Gov’t, AI Opportunities Action Plan: One Year On (2026).
This is not necessarily wrong. A rigid statute can become obsolete before the ink dries. But the U.K. approach creates the same risk that appears in the American framework: AI governance becomes an adjunct of growth policy. The Vatican’s warning is that efficiency, economic growth and national capability are not neutral measures of progress. Progress must be measured by the dignity of each person and the good of all peoples, not merely by adoption curves, model performance, venture investment or public-sector productivity.
China: Control, Security and State Power
China’s AI governance is more interventionist and more explicitly tied to state security, social stability and information control. The Interim Measures for the Management of Generative Artificial Intelligence Services apply to public-facing generative AI services in mainland China and seek to promote development while safeguarding national security, public interest and social order. Interim Measures for the Management of Generative Artificial Intelligence Services, Cyberspace Admin. of China et al. (effective Aug. 15, 2023), translation here.
China’s model is the mirror image of the American model. The United States worries about overregulation impeding innovation and strategic leadership. China worries about uncontrolled information, social instability and threats to state authority. Both models treat AI as a strategic asset. The Vatican treats AI as a moral environment.
That difference is critical. Magnifica Humanitas warns that truthful information does not arise from “centralized or automated control,” but from verification, cross-checking, responsible argumentation and relationships of trust. Leo XIV, Magnifica Humanitas ¶ 132. That sentence is a critique of both platform capitalism and state censorship. Deepfakes, synthetic propaganda and algorithmic amplification can destroy democratic trust. So can state-managed truth.
The encyclical’s answer is not “let the market decide” and not “let the state decide.” It is an ecology of communication rooted in truth, human dignity, responsibility and recourse. That is difficult to implement. It is also exactly the problem that current AI policy has not solved.
Canada, Japan and South Korea: Middle Paths
Canada’s Artificial Intelligence and Data Act, originally tabled as part of Bill C-27, was designed to establish obligations for high-impact AI systems, although Canada’s legislative path has been uneven and politically contingent. Gov’t of Canada, The Artificial Intelligence and Data Act (AIDA) – Companion Document. Canada’s struggle illustrates a broader problem: Everyone wants trustworthy AI, but translating “trustworthy” into legislative obligations, institutional jurisdiction, enforcement capacity and innovation policy is hard.
Japan’s first AI-specific legislation, the Act on Promotion of Research and Development and Utilization of Artificial Intelligence-Related Technologies, passed in 2025, takes a promotion-oriented approach that sets national policy direction rather than EU-style prescriptive prohibitions. White & Case, Japan’s First AI Legislation Becomes Law (Apr. 14, 2026).
South Korea’s AI Basic Act, promulgated in January 2025 and entering into force in January 2026, creates a broader framework that combines promotion, transparency, labeling and obligations around high-impact AI. Ministry of Sci. & ICT, Republic of Korea, Korea’s AI Basic Act (Jan. 2025).
These middle-path regimes are important because they reject the false binary between deregulation and prohibition. They attempt to promote AI while managing risk. But the Vatican would still ask whether these frameworks adequately protect invisible workers, children, the poor, migrants, the elderly, the surveilled, the scored and the excluded.
The International Layer: Principles Without Police
At the international level, the OECD AI Principles, first adopted in 2019 and updated in 2024, promote trustworthy AI that respects human rights and democratic values. OECD, Recommendation of the Council on Artificial Intelligence, OECD/LEGAL/0449.
The United Nations General Assembly adopted a consensus resolution in 2024 on safe, secure and trustworthy AI systems for sustainable development. G.A. Res. 78/265, Seizing the Opportunities of Safe, Secure and Trustworthy Artificial Intelligence Systems for Sustainable Development (Mar. 21, 2024).
These efforts matter. They create diplomatic vocabulary, interoperability and soft-law pressure. But they are not law in the ordinary enforcement sense. They do not answer who audits frontier models, who has standing to challenge an algorithmic denial, who pays for harms, who protects data-labeling workers, who governs military AI, who limits biometric surveillance, or who prevents a few firms from becoming private sovereigns of digital cognition.
That is why Magnifica Humanitas uses the language of “disarming” AI. It does not mean banning AI. It means disarming the mentality of armed competition, monopolistic control and technical domination. It warns against the assumption that technical power confers a right to govern. Leo XIV, Magnifica Humanitas ¶ 110.
That sentence could have been written for Washington, Brussels, Beijing, London, Sacramento, Seoul and Silicon Valley.
What Governments Are Missing
Most government AI policy is built around four concepts: safety, competitiveness, transparency and accountability. Those are necessary. They are not sufficient.
Safety asks whether the model causes harm. Competitiveness asks whether the nation wins. Transparency asks whether affected parties can know something about the system. Accountability asks whether someone can be blamed or sued when things go wrong. The encyclical asks whether the system honors the person before anything goes wrong.
That shift changes the governance model. It means AI policy should not merely require post hoc impact assessments. It should require pre-deployment human-rights analysis, labor-chain review, environmental-cost accounting, cybersecurity testing, adversarial misuse analysis, appeal rights, data provenance, child-protection rules, procurement restrictions, and enforceable duties for developers and deployers. It means that “responsible AI” cannot be a brand statement. It must be a control framework.
For cybersecurity professionals, this means AI governance belongs inside enterprise risk management, not marketing, innovation or legal alone. The AI bill of materials should matter. Training data should matter. Red-team results should matter. Model access controls should matter. Prompt logging and retention should matter. Output monitoring should matter. Human override should matter. Vendor representations should matter. Incident response should include model compromise, data poisoning, prompt injection, synthetic identity fraud, deepfake-enabled business email compromise, and autonomous agent misuse.
For lawyers, the lesson is equally direct. AI governance is not only privacy law, not only IP law, not only employment law, not only product liability, not only civil rights, not only cybersecurity, and not only procurement. It is all of them. The organization that treats AI as a tool will miss the risk. The organization that treats AI as decision infrastructure has a chance.
The Bottom Line
The United States is trying to win the AI race. The European Union is trying to regulate the AI market. The United Kingdom is trying to accelerate AI adoption. China is trying to control AI’s social and political effects. Canada, Japan and South Korea are trying to find workable middle paths. International organizations are trying to create common principles.
The Vatican is doing something different. It is asking whether the race itself has become the problem.
That does not make Magnifica Humanitas anti-technology. Quite the opposite. It recognizes that technology can heal, connect, educate and protect. But it insists that technology becomes dangerous when it is governed by profit, domination, surveillance, disinformation or the fantasy that human weakness is a defect to be engineered away.
In cybersecurity terms, the encyclical is a threat model for the human person. The asset is dignity. The adversary is domination. The vulnerability is our willingness to trade freedom for efficiency. The exploit is convenience. The payload is dependence. The mitigation is not merely regulation, but governance rooted in truth, transparency, solidarity, subsidiarity and recourse.
The legal question is whether governments can convert those principles into enforceable rights and duties before AI becomes too embedded to challenge. The moral question is whether we are building Jerusalem or Babel.
Right now, much of the world is still buying bricks.
