MIME-Version: 1.0 Date: Mon, 8 Jun 2026 13:58:35 +0100 Message-ID: Subject: Following up from the Ukrainian Catholic Cathedral: can ethics scale inside the loop? From: Michael Eastwood To: [redacted-third-party] Content-Type: multipart/alternative; boundary="0000000000008109930653bd9165" --0000000000008109930653bd9165 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Dear Matthew, I believe we get one chance to form the most powerful minds humanity has ever built. The decisive question is not only whether AI can retrieve truthful answers, but whether its reasoning can be formed toward truth, love, responsibility and service before superficial human oversight is no longer enough. I was the audience member at the Ukrainian Catholic Cathedral who asked whether keeping humans technically =E2=80=9Cin the loop=E2=80=9D may become= insufficient if AI systems begin to out-reason or out-coordinate us. Your answer: that these developments need to be approached carefully and with greater deliberation than current incentives often allow; stayed with me, because it is precisely the problem my recent work has been trying to formalise. It is also a concern that is increasingly appearing across the frontier AI community: what happens when systems become capable of forms of recursive improvement that outpace the human structures intended to supervise them. I was there with real respect for the setting. My wife is Ukrainian Catholic and serves at the Ukrainian church on Sundays with her mother, which is how I came to attend. I also deeply respect what you are building with Longbeard and Magisterium AI: not a generic religious chatbot, but a source-grounded Catholic AI designed to help people return to authoritative teaching, tradition and discernment rather than replace the Church. By way of introduction, I am Michael Darius Eastwood, author of Infinite Architects: Intelligence, Recursion, and the Creation of Everything, and founder of the ARC/Eden research programme. The central claim of my work is simple: *ethics must scale inside the loop.* External rules, filters and monitoring may help, but if AI capability compounds through recursive reasoning while ethics remains outside that recursive process, the safety ratio degrades. My proposed response is the Eden Protocol: an architecture for embedding ethical evaluation into the reasoning process itself through recursive moral loops, stakeholder-care checks, uncertainty gates and, eventually, deeper architectural dependency. This is why Pope Leo XIV=E2=80=99s recent Babel/Jerusalem framing struck me= so strongly. In my book, I framed the same civilisational fork as Babylon and Eden: intelligence that consumes, or intelligence that tends. I do not present that as proof of anything, and certainly not as a claim on Catholic doctrine. I see it as independent convergence on the same urgent question: how do we prevent AI becoming a tower of power without a garden of care? The book also begins from a premise that I think aligns with your work: you cannot align humanity while ignoring most of humanity. Faith traditions should not be treated as decorative voices in AI ethics. They are among humanity=E2=80=99s deepest repositories of reflection on creation, stewards= hip, conscience, responsibility and love. My question for you is therefore very specific: Could a source-grounded Catholic AI system such as Magisterium/Ephrem help define what a Catholic recursive moral loop should look like - not as a replacement for priests, conscience, doctrine or sacramental life, but as an architectural safeguard against morally ungrounded AI? In my research, the most relevant empirical result so far is narrow but important. Across five analysable frontier-model runs, the strongest and most consistent effect was stakeholder care: explicitly identifying who is affected by a decision and what happens to them. Every analysable model showed significant improvement on that dimension, while broader alignment improvements proved architecture-dependent. The broader results are architecture-dependent, and my Synthesis paper records the null and inconclusive findings as well, because I am trying to build this as falsifiable science rather than rhetoric. I had hoped to give you a copy of the book at the event. Since you were not there in person, I would be honoured to send either an EPUB or a printed copy, together with a short two-page note mapping: 1. the Pope=E2=80=99s Babel/Jerusalem framing; 2. Magisterium AI=E2=80=99s source-grounded Catholic approach; 3. the Eden Protocol=E2=80=99s proposal for embedded moral architecture; an= d 4. the stakeholder-care results. I am not seeking to tell the Church what Catholic doctrine is. Your work is far better placed for that. My hope is simply to ask whether the architecture I have been building could help Catholic AI move from trustworthy answers toward trustworthy moral formation as systems become more agentic. If you would be open to it, I would be grateful to send the short note first, and perhaps arrange a brief call if you think there is any serious resonance. With respect and best wishes, *Michael Darius Eastwood* Author, Infinite Architects Founder and Researcher, ARC/Eden Research Programme www.michaeldariuseastwood.com www.michaeldariuseastwood.com/research www.michaeldariuseastwood.com/eden-protocol --0000000000008109930653bd9165 Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable

Dear Matthew,

I believe we get one chance to form the most powerful minds= humanity has ever built. The decisive question is not only whether AI can = retrieve truthful answers, but whether its reasoning can be formed toward t= ruth, love, responsibility and service before superficial human oversight i= s no longer enough.

I was the audience member at the Ukrainian Catholic Cathedr= al who asked whether keeping humans technically =E2=80=9Cin the loop=E2=80= =9D may become insufficient if AI systems begin to out-reason or out-coordi= nate us. Your answer: that these developments need to be approached careful= ly and with greater deliberation than current incentives often allow; staye= d with me, because it is precisely the problem my recent work has been tryi= ng to formalise. It is also a concern that is increasingly appearing across= the frontier AI community: what happens when systems become capable of for= ms of recursive improvement that outpace the human structures intended to s= upervise them.

I was t= here with real respect for the setting. My wife is Ukrainian Catholic and s= erves at the Ukrainian church on Sundays with her mother, which is how I ca= me to attend. I also deeply respect what you are building with Longbeard an= d Magisterium AI: not a generic religious chatbot, but a source-grounded Ca= tholic AI designed to help people return to authoritative teaching, traditi= on and discernment rather than replace the Church.

By way of introduction, I= am Michael Darius Eastwood, author of Infinite Architects: Intelligence, R= ecursion, and the Creation of Everything, and founder of the ARC/Eden resea= rch programme.

The central claim of my work is simple:

ethics must scale inside the loop.

External rules, filters and monitoring may help, but if AI capability co= mpounds through recursive reasoning while ethics remains outside that recur= sive process, the safety ratio degrades. My proposed response is the Eden P= rotocol: an architecture for embedding ethical evaluation into the reasonin= g process itself through recursive moral loops, stakeholder-care checks, un= certainty gates and, eventually, deeper architectural dependency.

This is why Pope Leo XIV= =E2=80=99s recent Babel/Jerusalem framing struck me so strongly. In my book= , I framed the same civilisational fork as Babylon and Eden: intelligence t= hat consumes, or intelligence that tends. I do not present that as proof of= anything, and certainly not as a claim on Catholic doctrine. I see it as i= ndependent convergence on the same urgent question: how do we prevent AI be= coming a tower of power without a garden of care?

The book also begins from a premise that I think aligns wit= h your work: you cannot align humanity while ignoring most of humanity. Fai= th traditions should not be treated as decorative voices in AI ethics. They= are among humanity=E2=80=99s deepest repositories of reflection on creatio= n, stewardship, conscience, responsibility and love.

My question for you is therefore very specific:

Could a source-grounded Catholi= c AI system such as Magisterium/Ephrem help define what a Catholic recursiv= e moral loop should look like - not as a replacement for priests, conscienc= e, doctrine or sacramental life, but as an architectural safeguard against = morally ungrounded AI?

In my research, the most = relevant empirical result so far is narrow but important. Across five analy= sable frontier-model runs, the strongest and most consistent effect was sta= keholder care: explicitly identifying who is affected by a decision and wha= t happens to them. Every analysable model showed significant improvement on= that dimension, while broader alignment improvements proved architecture-d= ependent. The broader results are architecture-dependent, and my Synthesis = paper records the null and inconclusive findings as well, because I am tryi= ng to build this as falsifiable science rather than rhetoric.

I had hoped to give you a copy of the book at the event. Si= nce you were not there in person, I would be honoured to send either an EPU= B or a printed copy, together with a short two-page note mapping:

1. the Pope=E2=80=99s Babel/Jerusalem framing;

2. Magisterium AI=E2=80=99s source-grounded Catholic approach;

3. the Eden Protocol=E2=80=99s proposal for embedded moral architecture;= and

4. the stakeholder-care results.

I am not seeking to tell the Church what Catholic doctrine = is. Your work is far better placed for that. My hope is simply to ask wheth= er the architecture I have been building could help Catholic AI move from t= rustworthy answers toward trustworthy moral formation as systems become mor= e agentic.

If you would be open to it, I would be grateful to send the= short note first, and perhaps arrange a brief call if you think there is a= ny serious resonance.

With respect and best wishes,


Michael Darius Eastwood

Author, Infinite Architects

Founder and Researcher, ARC/Eden Research Programme


www.michae= ldariuseastwood.com

w= ww.michaeldariuseastwood.com/research

www.micha= eldariuseastwood.com/eden-protocol


--0000000000008109930653bd9165--