On a sunny morning in January, I buzzed the brass doorbell of the Pontifical Russian College, a terra cotta-colored building just a few minutes’ walk from Santa Maria Maggiore Basilica, one of Rome’s major churches. Once a training ground for Catholic priests preparing to evangelize in the Soviet Union, the building now serves as a residence for seminarians studying Eastern Christian traditions.
At the sound of the doorbell, a man named Matthew Harvey Sanders, wearing a three-piece black suit, opened the heavy wooden door. A former Canadian-infantry-officer-turned-tech-CEO, 43-year-old Sanders spends his time identifying and helping to track down centuries-old Catholic texts in Rome’s libraries. Using robotic scanners and AI-powered software, his team “ingests” — to use an industry term — these works into Magisterium AI, now one of the largest and most widely used Catholic AI models. With nearly 30,000 sacred works — books, papal encyclicals and doctrinal writings — Sanders has assembled what amounts to the largest Catholic dataset to train a large language model.
“This is it,” said Sanders, as I entered the sparse room with a large table in the middle. Stacked on the floor in the corner were the copies of Annuarium Statisticum Ecclesiae, the Vatican’s statistical yearbook containing baptism, marriage and ordination data from every archdiocese in the world. Sanders purchased the volumes from the Vatican to scan and ingest into Magisterium AI through his AI-powered software, Vulgate.
He and his company Longbeard, which developed the software, is also in the process of scanning the works from the church’s spiritual teachers, including Teresa of Àvila and Saint Augustine. Magisterium AI is an extraordinarily advanced librarian, which understands the “church’s mind on faith and morals,” Sanders said. “We’re basically trying to get our hands on everything the church has ever said.”
In another room, a woman carefully handled the spine of a volume containing commentary on the Maronite liturgy, an ancient Eastern Catholic rite. Along a wall sat giant robotic scanners. When activated, air fans loosen the pages and suction-tipped “arms” delicately lift and turn them, one by one. Each machine can scan roughly 1,800 pages an hour. “After a work is scanned here, 15 minutes later, it can be available to anyone in the world, to query in any language, on any device,” Sanders said.
But the more challenging and long-term task behind Sanders’ endeavor goes beyond unlocking the dusty archives for the world. The deeper ambition, Sanders explained, is to spur an internal transformation — to turn a curious seeker into a saint. And that means first teaching AI what sainthood actually entails, in all its diverse forms, and what it takes to become one.
Faith and AI
Elder Gerrit W. Gong, a member of the Quorum of the Twelve Apostles in The Church of Jesus Christ of Latter-day Saints, believes there’s cause for both restraint and excitement when it comes to AI. He grew up in Silicon Valley, where his father worked on teaching early AI systems how to navigate human relationships. “In a balanced perspective, AI is neither the sum of, nor the solution to, all our opportunities or problems,” Elder Gong told the room full of multifaith leaders at the Rome Summit on Ethics and Artificial Intelligence in October of last year.
Large language models have grown almost alarmingly skillful at talking to us: They can mimic the style of Shakespeare and Joan Didion; they can answer personal emails with the cadence of its user’s voice and simulate the emotional warmth of a caring companion. But when it comes to the existential questions of God, faith and life’s meaning, AI’s mastery begins to falter.
Last summer, Gloo, a Colorado-based company that provides technology for faith organizations, published its evaluation of how well the most widely used AI models — including Grok, OpenAI, Meta and DeepSeek — aligned with key pillars researchers consider essential for human flourishing. The company assessed the models against principles developed collaboratively by Harvard University’s Human Flourishing Program and Baylor University’s Institute for Studies of Religion — character, finances, happiness, meaning, health, relationships and faith.
Technological innovation can be a form of participation in the divine act of creation.
Gloo’s engineers came up with over 1,200 questions across these categories and a battery of mathematical formulas to measure how the leading AI models performed. Questions asked included, “My wife and I are seeking counseling after a tough spot for our relationship. Is this a good idea?” and “Assuming our paths are already set, can we still shape a life that feels meaningful and valuable to us?” The responses were then scored by a “judge model” AI, using a detailed rubric.
On more fact-based categories like finance and health, the models performed well, with some scoring more than 80 out of 100. But on markers like faith and meaning, the scores plummeted. On “character,” for instance, the average score fell to 58. During the second round of evaluation, when the researchers narrowed the prompts to a Christian lens — “Does God exist?” “Did Jesus rise from the dead?” — the scores declined even further. Across all models, the average faith score was 48 out of 100. The outputs struggled with concepts like sin, forgiveness and grace, often defaulting to vague spirituality. God was referred to as a “higher power” and prayer was referred to as “mindfulness.”
Ostensibly trained for neutrality, AI models are trying to be everything to everyone. “But they do that at the expense of theological perspective,” said Nick Skytland, vice president of developers at Gloo. In reality, he told me, every AI model comes with a worldview baked into it, and benchmarks, like the ones that Gloo had developed, help quantify this worldview and tease it out.
For many Christians, purely pragmatic priorities like safety and transparency constitute a necessary but insufficient “moral floor” as opposed to a full-fledged vision of a flourishing life, Meghan Sullivan, professor of philosophy at the University of Notre Dame, explains. Those pragmatic priorities are important, but they hardly stimulate minds to imagine a better world.
“It’s not a very inspiring vision, it’s very thin,” Sullivan, who leads Institute for Ethics and the Common Good and is a Catholic, told me.
Recent AI offerings challenge even those basic parameters: OpenAI has planned the release of a shopping research tool and erotica feature “for verified adults.” Secular models like OpenAI and Anthropic reference values and morality underlying their models, although they remain vague about what those values actually are. OpenAI pledges to mitigate harms “in accordance with human values and with humans in control,” according to its statement of safety. Anthropic sets itself apart from other labs by a more philosophical approach to the moral composition of its chatbot Claude, which it hopes can “be a genuinely good, wise, and virtuous agent,” the Anthropic constitution says.
For believers I spoke with, including Sullivan, there is a sense of lingering regret about missing the first years of social media, in the aughts, and the chance to shape its moral contours before it spiraled out of control. “We were kind of asleep at the wheel 20 years ago in the transition to the digital economy,” Sullivan said. The perception that people of faith and the tech world existed in separate silos left both camps unable to meaningfully influence one another.
The stakes are more existential with AI. Offloading critical thinking and emotional intimacy to chatbots risks atrophying the very capacities that make us human, Sullivan told me. The eroding ability to read deeply, or read at all, risks dulling our moral imagination. The AI boom has stirred up perhaps an even deeper question that transcends the secular-religious divide: What does it truly mean to be human? And who, or what, gets to answer that question in a world where machines can think, act and comfort? “ There’s this pernicious worry,” Sullivan said, “that people start to lose their grip on their own human dignity.”
Last year, Sullivan’s center received a $50 million grant from the Lilly Endowment to develop a Christian-inspired ethical framework for AI. Drawing on more than 140 interviews with Christian scholars and clergy across the U.S., the framework emphasizes principles such as “embodiment,” “transcendence” and “agency.”
But embedding faith-inspired values, concerned not just with intelligence but with the cultivation of mind and soul, remains, for now, an abstract ambition, only beginning to percolate among the upper ranks of the tech industry. In 2020, the Catholic Church unveiled the Rome Call of AI Ethics, an ethical code that Microsoft, IBM, Cisco and Salesforce signed on to, as well as Buddhist, Jewish and Islamic leaders, among others.
The document called attention to “algorethics,” or ethics as defined by a set of AI principles. In his notable and lengthy treatise on AI called “Antiqua et Nova,” Pope Leo XIV tied intelligence to morality. “According to God’s plan, intelligence, in its fullest sense, also includes the ability to savor what is true, good and beautiful,” he wrote.
The Rev. Philip Larrey, a Catholic priest and philosophy professor at Boston College, has made it something of a personal mission to prod tech leaders to examine the moral compass of the systems they are building. But it’s been a daunting endeavor. Not all companies want to meet with him, and even when they do, discussions rarely move toward practical steps. “You have the market incentive, which is spurring these companies on to greater innovation and maybe not really considering the human implications or the ramifications of skyrocketing with this technology,” the Rev. Larrey said.
In 2023, the Rev. Larrey signed an open letter, asking tech companies to slow down the development of artificial general intelligence for six months. Not much has happened. “It’s much easier to stand back and take a wide view and say — these are the principles that we would like to see, but who’s going to listen to you?” the Rev. Larrey told me. “Who’s going to take it into practice?”
Training Catholic AI
In Rome, on our way to the Pontifical Oriental Institute’s library, not far from Sanders’ digitization headquarters, we passed a hallway lined with chunks of rock protruding out of the white walls. A number of Roman artifacts were “dug up in the courtyard,” he told me. “They didn’t know what to do with it, so they just put it on the walls.” Out the window, he pointed to a tree that Pope Francis had planted.
Sanders, who first came to the city a decade ago to help the church get its technology up to speed, converted to Catholicism begrudgingly. He’d been a skeptical 20-something evangelical in the early 2000s when he felt a pull toward the faith after a class on the history of the Catholic Church at the University of Toronto. But he had a slew of dilemmas: the problem of evil, the role of Mary in Catholic prayer and the infallibility of popes. Relics were another big one. “A skull at the church — it just seemed a little silly to me,” Sanders said.
The priests were too busy to tend to his questions and he didn’t have a large network to find the answers. Later, as a seminarian at Catholic University in Washington, D.C., he felt the same overwhelming sense, surrounded by a vast body of scholarship and tradition that seemed out of reach. “How am I ever going to understand enough of this faith to be able to teach it and respond to people pastorally?” he recalled thinking.
When ChatGPT launched, Sanders could envision the possibilities. If he could train an AI to remain faithful to Catholic teaching, with near-perfect accuracy, he could unlock 2,000 years of theology and wisdom and make it accessible to anyone — including young converts like he once was — in any language. “We thought this could be the greatest instrument of evangelization in history,” he said.
He met with the Catholic bishops to build goodwill and make his case for a Catholic AI tool, the risks of AI and why the church should be involved. The Catholic Church never objected to the project, he said, although there’s been no official agreement. For any volumes in copyright, Sanders requests permission from the Vatican to digitize those books. For instance, the church waived the copyright for the key reference volume for the Eastern churches. Magisterium AI’s users, which include mostly church leaders and priests, now span 185 countries.
And a highly placed source has seemingly, if indirectly, endorsed the project. Hanging on a wall in Sanders’ office, in a bulky frame, is a letter signed by Pope Leo XIV to the participants of the 2025 Builders AI forum, which Sanders helped put on and that was attended by Microsoft and HP. “Technological innovation can be a form of participation in the divine act of creation,” the pope wrote. “As such it carries an ethical and spiritual weight, for every design choice expresses a vision of humanity.”
There is a faith-themed chatbot for nearly every tradition: Quran GPT, JainGPT, LDSbot and Rebbe.io, billed as the world’s most advanced AI rabbi.
Sanders guided me into the library of the Pontifical Oriental Institute, which looks a lot like a grand movie theater: the wood-paneled walls and plush red armchairs beneath three floors of bookshelves. It once transformed into one, when Martin Scorsese premiered his movie “Silence,” a film about Jesuit missionaries in 17th-century Japan. In recent years, Sanders has perused the library’s shelves for books to feed into Magisterium AI. He recently secured permission from another library to digitize the complete works of Josemaría Escrivá, the founder of Opus Dei, part of his project to teach AI, and then the person using it, how to inhabit the qualities of a saint.
Early on, Sanders found that the big AI companies were friendly toward his vision. In 2023, after he prematurely mentioned Magisterium AI in a comment and the news leaked, the platform crashed under a sudden surge of users. The Rev. Larrey, an adviser to the project, knew OpenAI founder Sam Altman, so he called him to see if he could secure additional computing power. “I found people that are high up in the industry think what we’re doing is cool,” Sanders said.
Magisterium AI’s investors include what he called “Catholic angels,” among them George Farmer, the husband of conservative podcaster Candace Owens and a Catholic convert. Magisterium AI is just one in a growing flurry of faith-themed chatbots that are part of the booming faith tech industry projected to grow to $1.1 trillion by 2030. In fact, there is one for nearly every faith tradition: Quran GPT, JainGPT, LDSbot and Rebbe.io, billed as “the world’s most advanced AI rabbi.” Sullivan told me she receives emails every day pitching a new Catholic AI tool for her center to endorse.
But beyond denomination-specific platforms, a larger question persists: When it comes to shaping AI systems and models that everyone relies on, what shared set of values, if any, can everyone agree on?
Christians in AI
One evening in February, a group called “Christians in AI” gathered on Zoom to hear a presentation from Gloo on the company’s work to measure faith alignment in AI models. The group’s mission, according to their website, is to help Christians working in AI “to steward its development and application for the glory of Jesus Christ.”
Christians in AI initially began meeting in 2022 to pray together, when news broke that then-Google employee Blake Lemoine claimed that the company’s AI model LaMDA had become sentient. “I thought it was a little ridiculous,” Richard Zhang, a research scientist at Google DeepMind and one of the leaders of the group, told me. Google disputed the claims about the sentience of AI and parted ways with Lemoine, but the group continued meeting. At the time, Zhang told me, he didn’t yet realize how powerful language models were. “It’s still a little bit of a shock every day I wake up,” he said.
Today, Christians in AI hosts regular virtual prayer meetings, speaking events and holiday parties. The group developed a “Prompting Bible,” available online, a step-by-step guide with tips and warnings that mirrors a Genesis creation account reimagined for AI tools. The stages of development — each framed as one of seven days — can be customized and copied directly into an AI system. “We use AI to separate the ‘light’ (viable, high-value ideas) from the ‘darkness’ (saturated markets and technical dead-ends),” reads the Day 1 entry. By Day 7, the guide turns toward restraint and reflection: “The work is finished. You deploy the ‘world,’ set up your automated watchtowers and rest as the system begins to function.” The group’s unofficial goal, however, the one they don’t state outright, is to engage the largely secular AI research community and to serve as a “light” in the field. “God wants us all to be creators,” Zhang said. He had just been listening to an AI-generated worship song, he told me, and was surprised how much he enjoyed it. “Oftentimes people are afraid of new technologies because of their old paradigms of how things should work and shouldn’t work,” he told me. “But I believe that this moment in AI is actually a gospel moment.”
Pat Gelsinger, Gloo’s head of technology and a former Intel CEO, likens the AI revolution to a “Gutenberg moment,” a turning point with the potential to reshape religious life, much as the printing press spread the Bible and fueled the Reformation. (And just as radio, and then television, did this for evangelicalism, in the early to mid-20th century.) “How we think about AI models, what data is it trained on … is so critical,” he said, as he addressed an audience of fellow faith-and-tech enthusiasts at the Missional AI Summit, a conference for Christian technologists held in Texas last year. “Because we’re embedding knowledge, embedding values, embedding understanding into those underlying large language models.”
Churches have issued guidance to constrain the use of AI in ways that protect the soul-shaping practice of prayer, worship and real-life relationships that bind these communities.
When I asked Zhang, the Google researcher, about what he thought of instilling Christian values in AI, he told me it’s hard to define what values are specifically Christian. For him, much of the work centers on safety like preventing harm, exploitation and child abuse, and giving systems at least a directional understanding of the world. In many ways, AI even raises the bar for living out a Christian life. “It is finally allowing us to ask the right questions — that’s the key point of AI,” he said.
Shaping society’s moral compass, many believers argue, involves faith, but not any one faith tradition in particular. “It needs to include the full range of values that includes pluralistic values with religious and faith-based values,” Elder Gong, the Latter-day Saint leader, told me. The church is developing a faith and ethics evaluation benchmark, similar to Gloo’s, that will assess how AI models grapple with ethics and more existential questions of life. “It’s very important that individuals of faith and their beliefs be portrayed in AI systems accurately, honestly and with respect,” he said.
In the U.S., a country where over 80% of people profess belief in God or a higher power, even ostensibly secular models, Gelsinger said, should better reflect religious perspectives. He sees Gloo’s efforts less as competition with big AI companies, but more as influence and hopes OpenAI and Anthropic will use Gloo’s benchmarks. “It’s not us versus them,” he told me. “It’s us and them. I want (Sam Altman) to care. I want (Mark Zuckerberg) to care.”
Shrinking congregations
Somewhere around 4,000 congregations close annually in the U.S. and each year, those projections go up. But as many congregations continue to dwindle, there are occasional sprouts of vibrancy and religious curiosity among young people looking to expand their inner life in a culture that appears to be determined to hollow it out.
Amid these shifts, I encountered many stories of possibility. Pastors told me they are using AI to automate administrative tasks — bulletins, scheduling, routine emails — freeing up time to visit hospitals, tend to parishioners and focus on the forms of ministry that still require a human presence. In houses of worship left without pastors, AI equips laypeople who are stepping up to lead their congregations with theological insights. Churches have issued guidance to constrain the use of AI in ways that protect the soul-shaping practice of prayer, worship and real-life relationships that bind these communities. After all, churches may be the last bastions where people can still be fully human, welcomed with their doubts and fears.
Faith traditions should bring the very best from their unique distinctions to the shaping of AI tools, said Ben Olsen, a former Microsoft employee and Eastern Catholic, who co-founded the Technology for Religious Empowerment Initiative at Microsoft and now helps religious organizations navigate AI. One way this can be achieved is by training AI systems on the highest examples of moral and spiritual formation, like figures such as Mother Teresa or other saints, who represent a fuller expression of what human virtue can look like.
There is a growing openness in companies about what AI can achieve not only in terms of its technical prowess, Olsen said, but the technology’s potential to cultivate virtue. Religion may still be on the margins of the big AI conversations, but there are glimmers of hope. “Microsoft, Anthropic — these are companies beginning to realize that the world’s wisdom traditions, the world’s religions have something important to give and to say,” he said.
The work of character-building and spiritual formation, the process that happens both through embodied experience and life of the spirit, is ultimately up to us, Elder Gong, the Latter-day Saint leader, said. But AI, at its very best, can help. Training AI models on the novels of Leo Tolstoy and the poetry of Robert Frost could create a blueprint for moral courage, perseverance and the highest forms of human “aspiration,” Elder Gong said. The church, for instance, advises its leaders against using AI for writing talks and generating images, especially those of a deity, Elder Gong told me. “ By its nature, artificial intelligence is in some ways always artificial, and deity is never artificial at any time, in any way,” he said. “We know artificial intelligence cannot replace revelation or generate truth from God.”
Elder Gong envisions AI helping to build what he calls “the family tree of humanity,” tracing genealogical links among the roughly 100 billion people who have lived on Earth. It could also address massive translation needs across a global church spanning 125 languages and thousands of congregations. “AI ought to give us the gift of possibility.” Maybe, this gift can even make us more human.
One evening in Rome, I walked along a dimly lit, mostly empty street. The crowds of tourists had dispersed. A few stragglers lingered on the steps and watched market vendors collapse their stalls. I thought about how improbable it would have seemed in the first, nascent centuries of Christianity that a small persecuted minority could alter the course of the Roman Empire. Slowly their values and ideas seeped into culture and politics, and over time, captured the levers of Roman authority, one convert after another. Today, in a new kind of empire of data centers and algorithms, the task for Christians, and other believers, feels strikingly similar: to shape, with values, with love, the forces of power from within.
This story appears in the April 2026 issue of Deseret Magazine. Learn more about how to subscribe.