Technical questions
- What can the system do?
- How accurately?
- How quickly?
- At what cost?
The framework
A framework for Catholic moral discernment of AI
A practical method for judging AI development by the human goods it serves, the power it creates, and the responsibilities it preserves.
Pope Leo XIV's Magnifica Humanitas places Babel and Jerusalem before us: the reach of our own power, and the shared work of building a home where everyone belongs. AI development is a human choice about purposes, power, relationships, participation, and responsibility. The child learning, the worker beginning a profession, and the patient awaiting care give that choice its human measure.
Read the Church documents informing this frameworkA working definition
What is a moral red line?
A moral red line sets a boundary where technological choices endanger a human good we have a duty to protect.
The six proposals take different forms: limits on development, restrictions on harmful design objectives, decisions reserved to responsible people, and conditions for effective oversight. Each also identifies a constructive direction.
Two levels of evaluation
AI must be judged by more than performance
Moral questions
- What human good does it serve?
- What authority does it redistribute?
- What habits and relationships does it form?
- Who bears the risks?
- Who remains responsible?
- Can affected people challenge or reverse its effects?
Capability tells us what a system can do. Moral discernment asks whether it should be built, deployed, or normalized.
Catholic foundations
Six Catholic criteria for judging AI
Intrinsic dignity
Human worth is never determined by intelligence, productivity, autonomy, or economic usefulness.
Truthful relationship
Technology should help people reach one another, protect trust, and make its artificial identity clear.
Responsible freedom
AI should expand people's ability to understand, choose, and act with their own judgment.
Common good & justice
Technological benefits and risks must be judged in relation to society as a whole, especially those with least power.
Solidarity & protection of the vulnerable
People affected by technological change must not bear its costs alone while benefits are concentrated elsewhere.
Subsidiarity & participation
Decisions should remain close to competent persons and communities, with affected people able to shape, challenge, and remedy them.
Magnifica Humanitas is the central interpretive foundation, read alongside Catholic teaching on dignity, conscience, education, work, solidarity, and participation. These principles ask how people can grow in freedom and care for one another through the technologies they build.
A six-step process
Discern upstream
Begin while choices about purpose, design, investment, and deployment are still open.
- 01
Purpose
What problem is the system intended to solve, and what objective is being optimized?
- 02
Human good
Which human capability, relationship, institution, or social good is affected?
- 03
Power
Who gains new authority, dependence, or control?
- 04
Formation
What habits, skills, expectations, or forms of dependence will repeated use create?
- 05
Participation
Who is affected, and do they have a meaningful voice before deployment?
- 06
Remedy
Can the system be challenged, corrected, stopped, or withdrawn when serious harm occurs?
Governance tests
The burden of justification rises with the stakes
Greater autonomy, opacity, scale, vulnerability, and irreversibility require stronger evidence and stronger human safeguards before deployment.
Evidence
Are the claimed benefits and safety conditions supported by credible evidence?
Control
Can competent people supervise, correct, override, and stop the system?
Accountability
Is responsibility clearly attributable to identifiable people and institutions?
Participation
Have affected persons and communities had a meaningful voice?
Recourse
Can decisions be explained, challenged, reconsidered, and remedied?
Public legitimacy
Has a decision with broad social consequences received appropriate democratic or institutional authorization?
From method to action