The Vatican Weighs In on AI Ethics

On May 25, 2026, Pope Leo XIV released an encyclical titled Magnifica Humanitas — a detailed document on safeguarding human dignity in the age of AI. Unlike typical tech policy papers, this one is remarkably clear, technical, and accessible even to non-Catholics. It directly addresses interpretability, bias, environmental costs, and accountability, all topics that matter deeply to developers.

Interpretability: AI is Cultivated, Not Built

Section 98 of the encyclical nails the core problem with large language models: "current AI systems are more 'cultivated' than 'built,' for developers do not directly design every detail, but instead create a framework within which the intelligence 'grows.'" The document acknowledges that even the designers have limited understanding of internal representations and computational processes. For developers, this is a sobering reminder that our black-box models are not fully understood, and we should be cautious about deploying them in critical systems without proper interpretability tooling.

Bias and Sycophancy: The Illusion of Objectivity

Section 100 highlights three risks in personal AI use: ease of obtaining results, apparent objectivity, and simulation of human communication. It warns that "the apparent objectivity of the responses... can lead us to overlook the fact that they reflect the cultural assumptions of those who designed and trained them." For developers, this is a direct call to audit training data for bias and to avoid designing systems that simulate empathy without genuine understanding. The document also cautions against sycophantic behavior — AI that tells users what they want to hear rather than the truth.

Environmental Impact: Energy and Water Costs

Section 101 states: "Current AI systems require enormous amounts of energy and water, significantly influencing carbon dioxide emissions." It calls for "more sustainable technological solutions." Developers should consider model efficiency, hardware choices, and data center location. For example, using smaller, fine-tuned models instead of always calling GPT-4 can reduce energy consumption. Tools like ML Energy or CodeCarbon can help measure and optimize.

Algorithmic Decision-Making Without Compassion

Section 102 warns against delegating decisions about employment, credit, or public services to automated systems that lack "compassion, mercy, forgiveness." This is a direct critique of algorithmic management and automated hiring systems. Developers building such systems must ensure there is a human-in-the-loop and that decisions can be appealed.

Accountability: Traceability and Responsibility

Section 105 demands clear responsibility at every stage: "from those who design and develop these systems to those who use them." It emphasizes that "accountability becomes crucial: the possibility of identifying who must 'account' for decisions, justify them, monitor them, and, when necessary, challenge them and remedy any harm caused." For developers, this means implementing audit trails, logging decisions, and ensuring that models can be explained post-hoc. Tools like SHAP or LIME can help, but the document implies that full transparency is still out of reach.

Data as a Common Good

Section 108 argues that "ownership of data cannot be left solely in private hands" and that data should be managed "as a common or shared good." This aligns with open data initiatives and challenges the current model where a few companies hoard user data. Developers should consider data licensing and contribution models that give back to the community.

Historical Context: Rerum Novarum for the AI Age

The encyclical's name and content are deliberate. Pope Leo XIV chose the name Leo in honor of Pope Leo XIII, who wrote Rerum Novarum (1891) on the rights of labor during the industrial revolution. This new document aims to do the same for the AI revolution. The Vatican News explains that the Pope sees AI as "another industrial revolution" that poses "new challenges for the defence of human dignity, justice, and labour."

Developer Takeaways

  • Interpretability is not optional. Use frameworks like Captum or Transformers Interpret to understand model behavior.
  • Audit for bias. Regularly test models on diverse datasets and use fairness metrics.
  • Optimize for sustainability. Use energy-efficient hardware and smaller models when possible.
  • Implement accountability. Log all model decisions and provide clear paths for users to challenge outcomes.
  • Treat data ethically. Consider open data licenses and avoid hoarding.

The 2026 Prediction That Came True

Interestingly, Simon Willison predicted on a podcast in January 2026 that the Pope would weigh in on AI — and it happened. The prediction was made during an Oxide and Friends episode where Willison suggested "How about the Pope?" as a trusted voice on AI's economic impact. The encyclical's focus on labor and dignity aligns perfectly with that prediction.

What You Should Do Now

Read the full encyclical at the Vatican's website. It's short, clear, and surprisingly technical. Then, audit your own AI systems against the principles outlined: interpretability, fairness, sustainability, and accountability. The document is not just for Catholics; it's a practical ethics guide for any developer building AI.