Ad maiorem Dei gloriam

Jesuit AI Ethics

A framework, a rubric, and a workshop · Catholic Social Teaching · the Rome Call · Ignatian discernment

Colophon

How this site was made

This site argues that AI must be used transparently, under human responsibility, and only insofar as it serves the mission. It was itself built with AI assistance. Those two sentences must be able to stand next to each other. This page is where they do.

I · Disclosure

What the machine did, and what the human did

This site was drafted, coded, and designed in July 2026 with substantial assistance from Claude (Anthropic's Fable 5 model), working under the direction of a human author. The AI wrote the prose and the code, fetched and checked the sources, and proposed the design. The human author defined the project and its standards, supplied the framework's intellectual claims, challenged and redirected the work, including rejecting an earlier design and asking for stricter verification, and reviewed and approved what was published. The author, not the tool, is responsible for every claim on this site.

The auditor tool works the same way in miniature. Text you submit is sent to Google's Gemini API to generate an assessment. It is not stored by this site, but it does leave this site, so do not paste confidential or personally identifying information.

II · Why this is consistent, not contradictory

The framework's conditions, applied to itself

A fair critic will ask whether a site about the dangers of AI may be built by AI. The framework's answer is that it teaches no doctrine of abstinence. It teaches conditions for right use, and it should be judged by whether its own making met them.

Human responsibility. Antiqua et Nova requires that “ultimate responsibility for decisions made using AI rests with the human decision-makers” (¶44). That holds here. The AI drafted. A named human authored, corrected, and answers for the result.

Transparency. The Rome Call's first principle, and the first thing this site's auditor checks, is honesty about AI's involvement. This page exists for that reason and is linked from every page's footer.

Truth held non-negotiable. Every quotation was verified against the primary documents before publication. The verification caught errors in the human-supplied working brief, and they were corrected rather than reproduced. The record is below.

Tantum quantum. The test of the Exercises, that created things be used insofar as they serve the end, was the working rule. The AI was used for drafting, verification, and construction, which is toolwork on an artifact. It was not used for what the framework warns about. No person was scored. No relationship was displaced. No judgment about a human being was delegated.

III · The verification record

What was checked, and what was corrected

Before publication, every quotation and citation was checked against its primary source. The Rome Call principles were checked against the signed document itself, extracted from the official PDF, since the Rome Call website displays shortened summary wording. The permanent principles were checked against the Compendium (§160, with §§161–162 on their permanence and unity). Each papal document was checked at its Vatican source with paragraph numbers confirmed: the World Day of Peace 2024 message, Dilexit Nos, Antiqua et Nova, Dilexi Te, and Magnifica Humanitas.

The verification found and corrected errors in the working brief.

  • The brief's URL for the World Day of Peace 2024 message was dead. The correct Vatican URL is used and linked.
  • The brief said Magnifica Humanitas warns against humans seeing themselves as “probabilistic functions.” The encyclical does not use that phrase. Its actual language warns against translating “the mystery of the person” into “data and performance” (¶9). The accurate wording is used throughout.
  • The brief listed the “Italian government” among the Rome Call's original signatories. The signatory was the Italian Minister for Technological Innovation.
  • The brief omitted Antiqua et Nova and Dilexi Te entirely, and the former is the Church's most systematic document on AI. Both were added.

The record is published for a reason. A process that discloses its corrections can be examined on the ones it has not found yet. If you find an error, say so, and it will be corrected and recorded here.

IV · The site examined by its own rubric

Five questions, asked of ourselves

The rubric is only credible if we are willing to sit for our own examination. Here it is, concessions included.

i.Who benefits first?

The intended first beneficiaries are mission-driven institutions and, through them, the students and communities most exposed to institutional AI. The site is free, requires no login, carries no advertising, and links every claim to a primary source. The site also benefits its author, since scholarship and reputation are real interests. Pretending otherwise would fail this question's own test.

ii.What happens when it fails?

If the auditor's analysis service is unavailable, the tool says so plainly instead of failing silently. The framework, rubric, and sources stand on their own without it. If the content fails, through an inaccurate quotation or a wrong paragraph number, corrections are publicly invited. The verification record above shows that errors found so far were fixed and recorded rather than hidden.

iii.What data is collected and why?

No accounts and no advertising. Text submitted to the auditor or the workshop is sent to Google's Gemini API for the sole purpose of generating the assessment, draft, or examination, and this site does not store it. The workshop's working session is saved only in the user's own browser. The site counts page visits through Vercel Web Analytics, which is cookieless and aggregates anonymized counts; it does not identify visitors, and no text anyone submits is included. Two further qualifications apply. Our hosting provider keeps standard operational logs. And once text reaches a third-party AI service it is governed by that service's terms, which is why both tools advise against pasting confidential or personally identifying information.

iv.Who maintains it?

A single author, with AI assistance. That is a limitation. Single-maintainer projects can decay, and Church documents will keep coming. Two mitigations apply. Every quotation is linked to the primary document, so the sources outlive this site. And the verification record makes the site's claims checkable by anyone, not only its maintainer.

v.Does it connect or isolate?

The site is designed as a bridge to the primary documents and to human deliberation. The rubric is meant for committee tables. The auditor's output is the start of a discernment rather than a verdict that replaces one. A summary can become a substitute for reading, and a tool can become a substitute for judgment. The mitigation is structural. Sources are linked everywhere, and the auditor recommends conversations rather than closing them.

V · What remains open

Concessions, held as an Examen

Three critiques are not answered above, because they should not be answered quickly.

Our common home. Building this site, and every query its auditor runs, consumes energy and water in data centers. The use was judged proportionate to the end. That is a judgment rather than a receipt, and the Universal Apostolic Preferences oblige us to keep making it rather than make it once.

Labor and formation. No one's work was displaced by this small, single-author project. At institutional scale, that answer does not generalize, and this is why the framework asks who maintains a system and who benefits first. We claim no exemption from those questions.

The heart. Dilexit Nos asks whether an age of frictionless tools erodes the interior life that gives work its meaning. Whether a discipline of discernment fully answers that worry, we do not claim to know. We hold it as the Examen holds everything, as a question to return to rather than a problem marked solved.

A framework that cannot survive its own examination deserves to fail it. This page is our attempt to pass honestly, and a standing invitation to tell us where we have not.

VI · Cite this work

If this work serves yours

The framework, the rubric, and the workshop may be cited in committee documents, syllabi, policies, and scholarship. A citation is also how improvements find their way back: cited work gets read, and read work gets corrected.

APA

Patil, V. H. (2026). Jesuit AI Ethics: A framework, rubric, and workshop for AI deployment in mission-driven institutions [Web application]. https://jesuitaiethics.org

BibTeX

@misc{patil2026jesuitaiethics,
  author = {Patil, Vivek H.},
  title  = {Jesuit AI Ethics: A framework, rubric, and workshop for
            AI deployment in mission-driven institutions},
  year   = {2026},
  url    = {https://jesuitaiethics.org},
  note   = {Web application}
}

Consistent with the disclosure above, the site was built with AI assistance under the author's direction; responsibility for its claims rests with the author, which is what a citation names.