Under The Hood: Magnifica Humanitas

18/08/2026

I read Pope Leo XIV’s encyclical Magnifica Humanitas beside the United Nations, Singapore and Ministry for Regulation AI instruments, and argue that Te Mana Raraunga’s principles of Māori data sovereignty come closest to asking by what right, and answerable to whom, AI is governed.

Under the Hood is about public administration, data, information, analysis, institutions, artificial intelligence and the machinery beneath the slogans. This post is the detour I promised at the end of the last instalment: a reading of Pope Leo XIV’s encyclical Magnifica Humanitas, placed beside the governance instruments that governments are actually producing.

At the end of the last instalment, I said I had been struck by the force of Pope Leo XIV’s Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence: by its treatment of AI as a question of human dignity, its concern about concentrated power, and its insistence that technology remain subject to responsibility, public control, and the common good (Leo XIV, 2026).

I am not reading it as theology, and you do not need to share its theology to take it seriously. I am reading it the way I read the Ontario audit in the series introduction: as evidence of what a serious institution sees when it looks at the technology we are examining in this series (Office of the Auditor General of Ontario, 2026).

The encyclical was issued on 15 May, the 135th anniversary of Rerum Novarum, and the timing is not decorative: the Church understands AI as a question about new social and economic arrangements rather than as a technology question.

The document is long, some 245 paragraphs, and three of its arguments belong in our discussion.

The first concerns where power now sits. “In the past, it was largely up to the State to guide and direct innovation. Today, however, the main drivers of development are private, often transnational, parties that are endowed with resources and the capacity to intervene that surpass those of many Governments” (Leo XIV, 2026, para. 5). The encyclical then inverts subsidiarity, the old principle that decisions belong at the level closest to the people affected. In the digital realm, “the highest level is not the State, but rather major economic and technological actors that exercise de facto power over the conditions of everyday life,” and subsidiarity therefore requires of those actors what we normally require of governments: “independent checks, transparency regarding algorithms, equitable access to data and avenues for recourse” (para. 71).

The second concerns delegated judgement. Sensitive decisions “risk being fully delegated to automated systems” that do not know “compassion, mercy, forgiveness, and above all, the hope that people are able to change” (para. 102). When an algorithm is entrusted “in practice with the power to select who is worthy or not, without anyone bearing responsibility for that judgment,” then “the exclusion of the vulnerable becomes cloaked in a veneer of neutrality and objectivity, against which it becomes difficult to raise objections” (para. 103). That passage describes, with some precision, the pattern this series has already traced through Robodebt and Horizon (Carney, 2019; Post Office Horizon IT Inquiry, 2025).

The third distinguishes the document from nearly everything else in the genre. The encyclical makes the point that ethics in the abstract is not sufficient: what is required is “robust legal frameworks, independent oversight, informed users and a political system that does not abdicate its responsibility” (para. 106). And then: “We cannot be satisfied with merely calling for the moralization of machines — the so-called ‘alignment’ of AI with human values… A more moral AI is not enough if that morality is determined by a few” (para. 107). This is a statement about ethics and power, and in particular about how the powerful ought to be ethical. The question it raises is who decided those values, in what forum, and with what opportunity for the rest of us to object, rather than whether machines can be given values.

The encyclical therefore asks three questions of any AI governance arrangement: where does the power actually sit, who bears responsibility for the judgement embedded in the machinery, and whose values are being operationalised.

Here is what struck me most, though, reading it with two chapters of the Oxford Handbook of Public Management this week, both published in 2007, neither mentioning AI: these are not new questions. The chapter on public management as ethics makes the responsibility question the organising problem of the entire field: “ethical action in public management plays out in discretion” (Dobel, 2007, p. 160), and whoever exercises that discretion carries “a degree of ineluctable personal responsibility in decisions and outcomes” (p. 162). He even saw the delegation problem coming, doubting that “contracts will capture the range of ethical subtleties and responsibilities that public servants address in decisions” (p. 172); substitute models for contracts and you are within reach of the encyclical’s paragraph 105. The chapter on virtual organisations speaks to the power question: e-government was already “replacing bureaucracy with information systems” (Margetts, 2007, p. 311), agencies were already “virtual in the sense that they cannot operate without the successful functioning of a major partnership” (p. 315), and she warned of a virtual state in which “accountability runs through the contract,” a form “furthest removed from democratic control” (p. 322).

The encyclical’s questions, in other words, are public administration questions that the AI governance discourse keeps forgetting it inherited. And I should be plain about the lens I am using, because it shapes everything that follows. I am not reading these instruments as a technologist, and I make no claim to be one. I come at this through political theory, through questions of consent, ethics, legitimacy, and good governance, and through the empirical field of public administration: how institutions actually behave, where discretion actually sits, and what happens to the people on the receiving end of the state. Those are also the encyclical’s questions, arrived at from a much older tradition of thinking about power and the human person, which is part of why I take it seriously as another long-lived institution reasoning from first principles to the same ground, rather than as a late arrival to the AI debate. Both discourses give me a way to analyse the AI governance instruments that follow.

The United Nations instruments

The UN’s Governing AI for Humanity report remains the fullest official map of the global governance deficit, and its recommendations have become institutions: the Scientific Panel appointed in February, and the Global Dialogue that met in Geneva this week (UN AI Advisory Body, 2024; United Nations, 2026). Given the current state of multilateralism, that is an achievement, and the Panel’s first report is a sobering document: AI capabilities are moving faster than any government’s ability to regulate them, and no technical guarantee of safety currently exists (Independent International Scientific Panel on Artificial Intelligence, 2026).

It is worth being honest, though, about what the machinery is made of: a panel that reports, and a dialogue that convenes. Geneva produced warnings and priorities rather than obligations. The UN instruments share the encyclical’s diagnosis of concentrated capability. What they cannot share is its bluntness, because bluntness is one of the things consensus documents trade away. The encyclical can say that a handful of actors must not “dictate these processes on their own” (Leo XIV, 2026, para. 72); a negotiated text arrives at the same idea through the gentler language of inclusion and capacity-building, and the obligation falls away in the translation. Pollyanna policy is not only a national genre.

The Singapore approach

Singapore is the opposite case. Whatever criticisms one might make of its approach, unimplementability is not among them. The Model AI Governance Framework has been iterated since 2019, a generative AI version followed in 2024, and in January Singapore produced what is described as the first framework anywhere for agentic AI, systems that act as well as recommend (Ministry of Digital Development and Information [MDDI], 2026). It is organised around the things Aotearoa’s Public Service AI Framework leaves largely to hope (Government Chief Digital Officer, 2025): bounding risk before deployment, making identified humans accountable, and, through the AI Verify toolkit, an actual means of testing systems against the stated principles.

If the last post argued that governance requires firm definitional edges, Singapore is what firm edges look like in practice. And yet, read after the encyclical, something is conspicuously absent. The Singapore instruments take the deployer’s objectives as given and ask how to pursue them safely. They do not ask the paragraph 107 question the encyclical calls attention to: whose values are in the machine, and who is able to contest them. I offer this as an observation rather than a criticism of Singapore on its own terms: operational maturity and democratic adequacy are different achievements, and it is possible to have one without the other.

The Ministry for Regulation guidance

Which brings me home. In the piece that started this series, Barb and I argued that the Public Service AI Framework names the right principles without the armour to convert them into implementable obligations (Te Kawa & Allen, 2026). The Ministry for Regulation’s Responsible AI in action is a different kind of document, and it comes closer than anything else the kāwanatanga side of the ledger has yet produced.

It begins from the regulatory context rather than the technology, starts deliberately with low-risk uses, triaging cases, finding patterns, validating data against clear rules, and counsels agencies to start small and build confidence before expanding. And it draws the line the encyclical draws, in language nearly as plain: “AI cannot interpret legislation, weigh fairness, or take responsibility for decisions. Those responsibilities must always remain with people” (Ministry for Regulation, 2026). That is, in the idiom of regulatory practice, the substance of paragraph 102: a non-delegable core of judgement, legal interpretation, and accountability, named without hedging.

Its limits are the familiar ones. It is guidance rather than obligation. Codes, in Dobel’s terms, “give rhetorical leverage and distill the essence of the values” (2007, p. 170), and that is not nothing; but what sustains values in practice is less a matter of “devices” than of “culture and ethos that socialize values and practice” (p. 164), and culture is precisely what guidance alone cannot supply.

And I want to acknowledge something the guidance does that most official instruments have not. It engages the data sovereignty discourse directly: it states that Māori data is taonga, that Te Tiriti o Waitangi obligations do not change when AI is involved, and it points regulators to Te Kāhui Raraunga’s Māori Data Governance Model (Ministry for Regulation, 2026). That deserves credit, and it is part of why, in my view, this document comes closest among the official instruments surveyed in this post.

The instrument that gets closest of all

Readers of the series introduction will know where this is heading, because I said there that any account of public sector governance in Aotearoa that fails to take Te Tiriti o Waitangi seriously is, by definition, incomplete, and that the data sovereignty discourse gives us a unique opportunity to shape the global conversation.

The encyclical is evidence for that claim, because the document in Aotearoa that comes nearest to Rome’s position is Te Mana Raraunga’s principles of Māori data sovereignty, which is not a government instrument at all (Kukutai & Taylor, 2016; Te Mana Raraunga, 2018).

Consider the correspondence. The encyclical argues that data “is the product of many contributors and should not be treated as something to be sold off or entrusted to a select few” (para. 108). Te Mana Raraunga begins from the proposition that data about Māori is a taonga over which rangatiratanga is exercised: that authority over data belongs with the people whose relationships the data carries. The encyclical calls for protecting “the ability of communities to make choices and corrections” (para. 109). Whakapapa, consent, collective benefit, and control articulate that same demand, and articulated it eight years earlier. Both begin with a prior question of authority and autonomy rather than with risk management. Both decline the premise, shared by every other instrument surveyed here, that governance is something done to data and people by whoever happens to hold power over the system.

The two positions are, of course, not identical. Te Mana Raraunga grounds authority in whakapapa and in Te Tiriti o Waitangi, in specific relationships between specific peoples and specific taonga; the encyclical universalises from human dignity, and a common good is not the same thing as a taonga under rangatiratanga.

But both Rome and Te Mana Raraunga are asking the same kind of question, and it is the question every other instrument declines to ask: by what right, and answerable to whom, as well as how we use this new technology safely and proportionately.

What this detour gives us

Placed side by side, each instrument is strongest where the others are weakest. The UN instruments name the concentration of power in the few but offer little in the way of binding that power, or holding it accountable. The Singapore instruments bind practice but do not reach the question of power. The Ministry for Regulation guidance names the non-delegable human core, further than any other official instrument in Aotearoa has gone, and that is welcome. But the data sovereignty principles go furthest of all, because they alone begin with questions of authority and accountability of the state.

That is the point I most want to leave with you. When one of the oldest and largest religious institutions in the world arrives, in 2026, at questions of authority over data that Aotearoa’s Te Mana Raraunga articulated as principle in 2018, that is an insight rather than an accident, and it confirms that Aotearoa already holds intellectual resources the rest of the world is only now reaching for, and that the failure of many official instruments to draw on them is an equity problem and, in the case of Aotearoa, a squandered advantage as well. The gap in those instruments lies in the prior question of who the principles answer to, rather than in the detail of the principles: a question our own field knew how to ask before the technology had its current name (Dobel, 2007; Margetts, 2007), and one the next instalment takes up directly, through the thinking and work of Te Mana Raraunga.

As ever, these are my evolving thoughts, not settled positions, written in my spare time. If you have read Magnifica Humanitas differently, or think I have been too generous to the Ministry for Regulation, or not generous enough, I would be glad to hear it.

References

Carney, T. (2019). Robo-debt illegality: The seven veils of failed guarantees of the rule of law? Alternative Law Journal, 44(1), 4–10. https://doi.org/10.1177/1037969X18815913

Dobel, J. P. (2007). Public management as ethics. In E. Ferlie, L. E. Lynn Jr., & C. Pollitt (Eds.), The Oxford handbook of public management (pp. 156–181). Oxford University Press.

Government Chief Digital Officer. (2025). Public Service AI Framework. Digital.govt.nz. https://www.digital.govt.nz/standards-and-guidance/technology-and-architecture/artificial-intelligence/public-service-artificial-intelligence-framework

Independent International Scientific Panel on Artificial Intelligence. (2026). First annual report. United Nations. https://www.un.org/global-dialogue-ai-governance/en

Kukutai, T., & Taylor, J. (Eds.). (2016). Indigenous data sovereignty: Toward an agenda. ANU Press. https://doi.org/10.22459/CAEPR38.11.2016

Leo XIV. (2026). Magnifica Humanitas: On safeguarding the human person in the time of artificial intelligence. Holy See. https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html

Margetts, H. (2007). Virtual organizations. In E. Ferlie, L. E. Lynn Jr., & C. Pollitt (Eds.), The Oxford handbook of public management (pp. 305–325). Oxford University Press.

Ministry for Regulation. (2026). Responsible AI in action: AI guidance for regulators. New Zealand Government. https://www.regulation.govt.nz/for-regulators/resources-for-regulators/ai-guidance-for-regulators/

Ministry of Digital Development and Information. (2026, January 22). Singapore launches new Model AI Governance Framework for Agentic AI [Press release]. https://www.mddi.gov.sg/newsroom/singapore-launches-new-model-ai-governance-framework-for-agentic-ai–/

Office of the Auditor General of Ontario. (2026). Use of artificial intelligence in the Ontario government: Performance audit (Special Report 2026). King’s Printer for Ontario.

Post Office Horizon IT Inquiry. (2025). Report: Volume 1. https://www.postofficehorizoninquiry.org.uk/volume-1-post-office-horizon-it-inquirys-final-report

Te Kawa, D., & Allen, B. (2026, May 11). Pollyanna policy: Is NZ’s framework for AI use in government overly optimistic? The Conversation. https://theconversation.com/polyanna-policy-is-nzs-framework-for-ai-use-in-government-overly-optimistic-281425

Te Mana Raraunga. (2018). Principles of Māori data sovereignty. https://www.temanararaunga.maori.nz/principles-of-maori-data-sovereignty

UN AI Advisory Body. (2024). Governing AI for humanity: Final report. United Nations. https://www.un.org/sites/un2.un.org/files/governing_ai_for_humanity_final_report_en.pdf

United Nations. (2026). Global Dialogue on AI Governance, Geneva, 6-7 July 2026. https://www.un.org/global-dialogue-ai-governance/en